Targeted bisulfite sequencing reveals changes in DNA methylation associated with nuclear reprogramming

Abstract

Current DNA methylation assays are limited in the flexibility and efficiency of characterizing a large number of genomic targets. We report a method to specifically capture an arbitrary subset of genomic targets for single-molecule bisulfite sequencing for digital quantification of DNA methylation at single-nucleotide resolution. A set of ~30,000 padlock probes was designed to assess methylation of ~66,000 CpG sites within 2,020 CpG islands on human chromosome 12, chromosome 20, and 34 selected regions. To investigate epigenetic differences associated with dedifferentiation, we compared methylation in three human fibroblast lines and eight human pluripotent stem cell lines. Chromosome-wide methylation patterns were similar among all lines studied, but cytosine methylation was slightly more prevalent in the pluripotent cells than in the fibroblasts. Induced pluripotent stem (iPS) cells appeared to display more methylation than embryonic stem cells. We found 288 regions methylated differently in fibroblasts and pluripotent cells. This targeted approach should be particularly useful for analyzing DNA methylation in large genomes.

Figure 1

Targeted bisulfite sequencing with padlock probes. (a) Each padlock probe has a common linker sequence flanked by two target-specific capturing arms (H1 and H2). H1 and H2 are melting temperature–normalized, and a spacer sequence is included to normalize probe lengths. The linker sequence contains priming sites (AP1 and AP2) for universal primers, two MmeI sites and a central AluI recognition site. (b) A CpG island (or other target regions) is covered by multiple padlock probes targeting partially overlapped regions on alternating strands. (c) A library of padlock probes is annealed to bisulfite-converted genomic DNA (black) and the 3′ ends are extended and ligated with the 5′ end. After removal of linear DNAs with exonucleases, all circularized padlock probes are PCR-amplified using a pair of common primers. (d) To generate a shotgun sequencing library, amplicons were reamplified in the presence of dUTP, digested with MmeI, and then with USER and S1 nuclease. Digested amplicons are end repaired and ligated with Solexa sequencing adaptors (green). Ligated products are then selected by size and amplified by PCR to generate the shotgun sequencing library. (e) Gel electrophoresis analysis of the padlock-captured products from two independent capturing reactions (1 and 2) and a no-template control. Expected amplicon size is in the range of 344–394 bp, which includes capturing targets (175–225 bp), capturing arms (58 bp) and amplification primers (111 bp). NTC, no-template control.

Figure 2

Normalization of padlock-capturing efficiency. (a) The ‘subsetting’ strategy. The 30,000 probes were divided into four sets (5 k, 10 k, 10 k, 5 k). The three less efficient sets were resynthesized. We reused the original 30,000-probe set because it was dominated by the most efficient 5,000 probes. (b) The ‘suppressor oligo’ strategy. (c) Distribution of normalized abundance for all captured targets with one 30,000-probe set and with four probe sets. The x-axis is the normalized abundance of each captured target, which is calculated by dividing the counts of the target by the average counts of all targets. The y-axis is the fraction of probes with the coverage equal to or greater than the normalized coverage. (d) Comparison of relative abundance for each target before and after normalization. The green vertical dash lines indicate the clear separation of four subsets of targets, as well as the fifth set normalized with the suppressor oligos.

Figure 3

Patterns of CpG methylation in fibroblasts and pluripotent cells. (a,b) Patterns of CpG island methylation on chromosomes 12 and 20 in twelve samples. Red indicates highly methylated CpG, and green indicates weakly methylated CpG. (c) Correlation between DNA methylation and gene expression. Each circle or square represents a 500-bp window. The windows in which average methylation levels are significantly correlated (P < 0.01) with gene expression are highlighted as orange circles. Gray squares indicate that the correlation is not significant. (d) Clustering of 26 selected genes based on the distribution of CpG methylation in the 4-kb TSS flanking regions. The 20 columns per cell line represent the fraction of probes within the TSS region where each probe exhibits a methylation frequency such that column header value ≤ methylation frequency < column header value + 0.05. For example, unmethylated TSS regions will have the highest probe fraction in the lowest value-labeled (left) column. Roughly half of the genes (14) were close to completely unmethylated in all cell types. Five genes (OCT4, ZIC3, NANOG, UTF1, DNMT3B) that show cell-type specific changes in methylation were grouped as a cluster. (e) Methylation pattern in the TSS flanking regions of OCT4. Two differentially methylated regions were marked by purple rectangles. (f) Hierarchical clustering of pluripotent cells and fibroblasts. The dissimilarity matrix was calculated based on Pearson correlation coefficients.