Characterizing the strand-specific distribution of non-CpG methylation in human pluripotent cells

Abstract

DNA methylation is an important defense and regulatory mechanism. In mammals, most DNA methylation occurs at CpG sites, and asymmetric non-CpG methylation has only been detected at appreciable levels in a few cell types. We are the first to systematically study the strand-specific distribution of non-CpG methylation. With the divide-and-compare strategy, we show that CHG and CHH methylation are not intrinsically different in human embryonic stem cells (ESCs) and induced pluripotent stem cells (iPSCs). We also find that non-CpG methylation is skewed between the two strands in introns, especially at intron boundaries and in highly expressed genes. Controlling for the proximal sequences of non-CpG sites, we show that the skew of non-CpG methylation in introns is mainly guided by sequence skew. By studying subgroups of transposable elements, we also found that non-CpG methylation is distributed in a strand-specific manner in both short interspersed nuclear elements (SINE) and long interspersed nuclear elements (LINE), but not in long terminal repeats (LTR). Finally, we show that on the antisense strand of Alus, a non-CpG site just downstream of the A-box is highly methylated. Together, the divide-and-compare strategy leads us to identify regions with strand-specific distributions of non-CpG methylation in humans.

Figure 1.

The ‘divide-and-compare’ strategy shows CHG and CHH methylations in human have no essential difference. (A) Schema of the ‘divide-and-compare’ strategy, (B) Spearman’s ranks correlation analysis between CHG and CHH methylation pairs shows the two non-CpG methylation patterns are not essentially different in human as in Arabidopsis. The corresponding pairs are xyChG and xyChH (x, y can be A, C, G or T; H indicates A, C and T are considered collectively). The x- and y-axes show methylation level ranks (1–48, from high to low) of xyChGs and xyChHs in the CHG list and CHH list, respectively. The rhos are calculated by Spearman’s ranks correlation test.

Figure 2.

Strand-specific non-CpG methylation in human gene body. (a) Diagram of the four regions analyzed in this study. Three intronic subregions are examined: 5′SS (5∼125 bp downstream of donor site), MI (120-bp region in the center of the intron) and 3′SS (125∼5 bp upstream of acceptor site). (b) Density plots of the skew scores for CpG methylation in different regions. x-axis, skew score. For skew score, 0 indicates no skewness, positive value indicates higher mCpH in antisense strand and negative value indicates the opposite. All the three subregions in intron showed significant skew of non-CpG methylation. P-value, one-tailed t-test. (c) Density plots of the skew score for CpH methylation in different regions. Similar with (b), (d) the skew score of non-CpG methylation in introns are positively correlated with the transcription levels (H1). X-axis, rank of transcription levels from high (left) to low (right). Y-axis shows the skew score. The skew score of each site is the average of 1000 transcripts. (e) Asymmetric sequences in intron guide skewed non-CpG methylation (H1). Context study showed asymmetric sequence guided the skew of non-CpG methylation. Analysis is carried on 3-mer patterns (NCH) in 3′SS of H1. The average methylation levels (mP, upper) of each pattern showed similar methylation levels on both strands. Patterns are ordered by average methylation level. The proportions of these 3-mers in sequence composition (cP, middle) showed asymmetric distribution of sequence. The contributions to methylation levels in the whole (coP = mP × cP, lower) of each 3-mer pattern are shown in bars. The enrichment of non-CpG methylation motifs on the antisense strand makes the higher methylation levels of non-CpG.

Figure 3.

Characterize the strand-specific non-CpG methylation in transposons. (a) The strand-specific methylation levels of CpG sites in different transposon groups in ADS-iPSC. Alu elements and MIR elements are two subgroups of SINE elements. Heights of bars and error bars are the means and standard deviations of the average methylation levels in each chromosome. All groups show concordant CpG methylation levels on two strands; (b) the strand-specific methylation levels of CpH sites in different transposon groups in ADS-iPSC. Asterisk indicates P < 0.01. P-value, two-tailed t-test, (c) the methylation propensity (average methylation level) of each 3-mer on two strands of Alu. Error bar, standard deviation of the average methylation levels in each chromosome, (d) the sequence frequencies of each 3-mer on the two strands of Alu elements, (e) the methylation propensity (average methylation level) of each 3-mer on two strands of LINEs, (f) the sequence frequencies of each 3-mer on the two strands of LINEs, (g) the 25 bp position from 5′ end of Alu (antisense strand) shows high non-CpG methylation levels. From the structure of Alu, the high methylated non-CpG position is right after A-box, which is known as binding site of Pol-III together with B-box. The highly methylated non-CpG site is in TACAG context.