Hypomethylated domain-enriched DNA motifs prepattern the accessible nucleosome organization in teleosts

Abstract

Background: Gene promoters in vertebrate genomes show distinct chromatin features such as stably positioned nucleosome array and DNA hypomethylation. The nucleosomes are known to have certain sequence preferences, and the prediction of nucleosome positioning from DNA sequence has been successful in some organisms such as yeast. However, at gene promoters where nucleosomes are much more stably positioned than in other regions, the sequence-based model has failed to work well, and sequence-independent mechanisms have been proposed.

Results: Using DNase I-seq in medaka embryos, we demonstrated that hypomethylated domains (HMDs) specifically possess accessible nucleosome organization with longer linkers, and we reassessed the DNA sequence preference for nucleosome positioning in these specific regions. Remarkably, we found with a supervised machine learning algorithm, k-mer SVM, that nucleosome positioning in HMDs is accurately predictable from DNA sequence alone. Specific short sequences (6-mers) that contribute to the prediction are specifically enriched in HMDs and distribute periodically with approximately 200-bp intervals which prepattern the position of accessible linkers. Surprisingly, the sequence preference of the nucleosome and linker in HMDs is opposite from that reported previously. Furthermore, the periodicity of specific motifs at hypomethylated promoters was conserved in zebrafish.

Conclusion: This study reveals strong link between nucleosome positioning and DNA sequence at vertebrate promoters, and we propose hypomethylated DNA-specific regulation of nucleosome positioning.

Keywords: DNA methylation; DNA sequence; Nucleosome positioning; Vertebrate.

Fig. 1

DNase I-seq detects accessible nucleosome linkers within HMDs. a A representative genome browser view of DNA methylation, HMDs, and DNase I-seq pattern (signals per million reads) in medaka blastula embryos. Vertical line height of DNA methylation track indicates the ratio of methylated CpG. Black boxes represent HMDs. b A close-up view of single HMDs in (a). c Average profiles of DNase I-seq signal, DNA methylation, nucleosome core, and TSS counts around the accessible nucleosome linkers at the HMD boundaries. Vertical green dashed lines indicate the position of nucleosome core estimated from MNase-seq data. The top schema shows the position of nucleosomes (green ovals) and methylated CpGs (orange circle)

Fig. 2

Nucleosome positioning in HMDs is predictable by k-mer SVM. a Examples of prediction of nucleosome linkers (DNase I accessible regions) by k-mer SVM in HMDs on chromosome 8. Dark purple indicates the score higher than 0, light purple, lower than 0. Pearson’s correlation and its P value between DNase I signal and SVM score for every 20 bp along the HMD are shown on the right. b A histogram of correlations for all HMDs on chromosome 8. Green and gray boxes represent the number of HMDs with and without significant correlation (P < 0.05), respectively. c A histogram of distances between top 10 SVM_DNaseI-weight 6-mers within HMDs. Distances shorter than 3 bp were excluded from the histogram

Fig. 3

Specific DNA 6-mers are enriched in accessible linkers. Average profiles of the SVM score, distribution frequency of 6-mers with highest SVM-weight (GCTAAC) and its reverse sequence (CAATCG), SNP rate, and Kaplan occupancy around the HMD boundaries. Vertical green dashed lines indicate the position of nucleosome core, and the top schema shows the position of nucleosomes (green ovals) and methylated CpGs (orange circle) (same as Fig. 1c)

Fig. 4

Specific 6-mers enrichment in linkers are unique to HMDs. a An example of prediction of HMDs by SVM_hypo. The DNA methylation ratio, HMDs at the medaka blastula stage, and SVM prediction score (dark purple indicates a score higher than −1, light purple, lower than −1) are shown. b Performance of classification of HMD versus methylated sequences on chromosome 8 by SVM_hypo. ROC curve and the area under the ROC curve (ROCauc) (top), and precision–recall curve (bottom) are shown. c Comparison of SVM-weights between SVM_hypo and SVM_DNaseI for all 6-mers (left), and boxplots shows the difference of SVM_hypo-weights between the top 20 SVM_DNaseI 6-mers and all 6-mers. P value was calculated using non-paired Wilcoxon test. d A schematic of nucleosome positioning and specific DNA motif distribution in presumptive HMDs

Fig. 5

HMD-specific DNA sequence preference for the nucleosome linker is conserved among teleosts. a, b Average DNA methylation, nucleosome position, and SVM_DNaseI score around TSSs in HMDs (left) and methylated regions (right). For both medaka (a) and zebrafish (b), SVM_DNaseI score was calculated using SVM trained by medaka data sets