MethylRAD: a simple and scalable method for genome-wide DNA methylation profiling using methylation-dependent restriction enzymes

Abstract

Characterization of dynamic DNA methylomes in diverse phylogenetic groups has attracted growing interest for a better understanding of the evolution of DNA methylation as well as its function and biological significance in eukaryotes. Sequencing-based methods are promising in fulfilling this task. However, none of the currently available methods offers the 'perfect solution', and they have limitations that prevent their application in the less studied phylogenetic groups. The recently discovered Mrr-like enzymes are appealing for new method development, owing to their ability to collect 32-bp methylated DNA fragments from the whole genome for high-throughput sequencing. Here, we have developed a simple and scalable DNA methylation profiling method (called MethylRAD) using Mrr-like enzymes. MethylRAD allows for de novo (reference-free) methylation analysis, extremely low DNA input (e.g. 1 ng) and adjustment of tag density, all of which are still unattainable for most widely used methylation profiling methods such as RRBS and MeDIP. We performed extensive analyses to validate the power and accuracy of our method in both model (plant Arabidopsis thaliana) and non-model (scallop Patinopecten yessoensis) species. We further demonstrated its great utility in identification of a gene (LPCAT1) that is potentially crucial for carotenoid accumulation in scallop adductor muscle. MethylRAD has several advantages over existing tools and fills a void in the current epigenomic toolkit by providing a universal tool that can be used for diverse research applications, e.g. from model to non-model species, from ordinary to precious samples and from small to large genomes, but at an affordable cost.

Keywords: DNA methylation; MRR-like enzyme; MethylRAD; epigenomics.

Figure 1.

Schematic overview of the procedure for MethylRAD library preparation. Genomic DNA is digested with the restriction enzyme FspEI, producing 32-bp fragments including four-base 3′ overhangs. Adaptors with compatible overhangs (NNNN) are ligated to each end of these fragments. Tag density can be adjusted using adaptors with selective overhangs (e.g. NNNG). The constructs are amplified and purified by gel extraction. Sample-specific barcodes are incorporated in each construct by PCR, and the products pooled for sequencing.

Figure 2.

(a) Technical reproducibility of MethylRAD for detection of different types of 32-bp FspEI sites and (b) corresponding sequencing depth correlation between replicate libraries. For two primary target sites (CCGG and CCWGG), high reproducibility is observed for both site discovery and site depth.

Figure 3.

(a,b) Genome-wide comparison of Arabidopsis methylation patterns inferred by MethylRAD and WGBS. For WGBS data, methylation patterns were generated using all CGs and non-CGs in the genome. Although MethylRAD captures only a fraction of CGs and non-CGs from the genome, it can infer genome-wide methylation patterns that resemble those obtained by WGBS at single-base resolution.

Figure 4.

(a) An exemplary chromosomal distribution of (a) methylated CCGG sites and (b) methylated CCWGG sites detected by MethylRAD, RTR-MethylRAD and WGBS. Each vertical bar represents a restriction site. For each panel, ‘all sites’ refers to all predicted unique CCGG or CCWGG sites in the genome that can be possibly methylated, whereas ‘sites targeted by RTR’ refers to all predicted unique RTR-targeted sites in the genome. A large majority of methylated target sites detected by WGBS are also detected by MethylRAD. Reduction of sites density is achieved in the RTR library which targets about one-eighth of all predicted CCGG and CCWGG sites in the genome.

Figure 5.

Detection and methylation quantification of RTR-targeted sites by RTR and standard libraries. RTR library captures the majority of methylated RTR-targeted sites with acceptable methylation quantification accuracy.

Figure 6.

Rarefaction analyses of standard and RTR libraries for the performance of target sites detection (a,b) and methylation quantification accuracy (c,d) at different sequencing scales. Resampling of standard and RTR libraries reveals saturation at approximately 5 and 3 million reads, respectively. High methylation quantification accuracy is maintained (r > 0.97) at the saturated sequencing depths.

Figure 7.

Comparison of the reference-based and de novo analytical approaches for target sites detection and methylation quantification (for one-to-one matched sites). The de novo approach creates a cluster-derived reference (CDR) from high-quality reads, which recaptures the majority of target sites detected by the reference-based approach. High methylation quantification accuracy (r = 0.98 for CCGG sites (a) and r = 0.94 for CCWGG sites (b)) is achieved in the de novo approach.

Figure 8.

Sequencing depth correlation of CCGG sites between replicate libraries prepared from same amount (a) or different amount (b) of input DNA. High reproducibility is observed for very low input levels (e.g. 1 ng).

Figure 9.

Sequencing depth correlation of CCWGG sites between replicate libraries prepared from same amount (a) or different amount (b) of input DNA. High reproducibility is observed for very low input levels (e.g. 1 ng).

Figure 10.

MethylRAD analysis of the epigenetic basis of carotenoid accumulation in scallop adductor muscle. (a) Genome-wide distribution of differentially methylated sites. The sites showing significant elevation of DNA methylation in orange or white muscle are labelled in red and blue, respectively. Gene name abbreviations are shown for the four most significant sites in the chromosome 8. (b) The gene structure of LPCAT1 and its associated MethylRAD tags. The differentially methylated sites are indicated by red tag names. (c) Comparison of methylation levels between two groups for each tag in the LPCAT1 gene. P-values are shown for three sites with significant methylation difference between the two groups. (d) Gene expression profiling of LPCAT1. Significantly higher expression of LPCAT1 is observed in the orange muscle group than the white muscle group.