Genome-wide identification of 5-methylcytosine sites in bacterial genomes by high-throughput sequencing of MspJI restriction fragments

Abstract

Single-molecule Real-Time (SMRT) sequencing can easily identify sites of N6-methyladenine and N4-methylcytosine within DNA sequences, but similar identification of 5-methylcytosine sites is not as straightforward. In prokaryotic DNA, methylation typically occurs within specific sequence contexts, or motifs, that are a property of the methyltransferases that "write" these epigenetic marks. We present here a straightforward, cost-effective alternative to both SMRT and bisulfite sequencing for the determination of prokaryotic 5-methylcytosine methylation motifs. The method, called MFRE-Seq, relies on excision and isolation of fully methylated fragments of predictable size using MspJI-Family Restriction Enzymes (MFREs), which depend on the presence of 5-methylcytosine for cleavage. We demonstrate that MFRE-Seq is compatible with both Illumina and Ion Torrent sequencing platforms and requires only a digestion step and simple column purification of size-selected digest fragments prior to standard library preparation procedures. We applied MFRE-Seq to numerous bacterial and archaeal genomic DNA preparations and successfully confirmed known motifs and identified novel ones. This method should be a useful complement to existing methodologies for studying prokaryotic methylomes and characterizing the contributing methyltransferases.

Fig 1. MFRE cleavage and formation of library inserts.

A) Recognition sites (blue) and cleavage positions (arrows) of three commercially available MFREs. B) Product of MspJI (recognition site in blue) cleavage of the fully methylated motif CCWGG (boxed), before and after end repair. The m5C residue on each strand is the 17^th base from the 3’ end.

Fig 2. Overview of MFRE-Seq.

Genomic DNA containing motifs that are fully methylated (red dots) or hemi-methylated (open red circles) is digested with one or more MFREs. Size selection enriches for the small fragments that result from MFRE cleavage of fully methylated sites, and sequencing libraries are prepared from these fragments (adapters in green). Sequence reads are then mined for motifs. The computational method for doing so described in this work involves binning reads by length, enriching for CCRM reads by base-filtering, aligning, and examining the base distribution at each position. Base distributions can also be represented as a sequence logo, as shown here.

Fig 3. Example of theoretical fragment types generated by MFRE digestion.

For simplicity, DNA is drawn as a single line, methylated motifs as colored dots, and cut sites on either side as triangles with color corresponding to that of the motif. Fragments were classified as one of six categories: “motif-cleaved” (when exactly cut, these are CCMD fragments), “interstitial” (regions between motif-cleaved fragments), “overlap-short” and “overlap-long” (created by cutting CCWGG sites less than 30 bp apart), “concatenated” (reads spanning an expected cut site, which most often consist of a motif-containing CCMD fragment joined to an interstitial fragment), and “other” (created by more complicated situations such as 3 or more clustered motifs). (A) Examples of motif-cleaved, interstitial, and concatenated fragments. (B) and (C) Examples of different types of overlap fragments, depending on whether any cleavage occurs between the two nearby motifs.

Fig 4. Diagnostic statistics from Illumina sequencing of FspEI-digested E. coli K-12 DHB4.

This strain is methylated by Dcm at CCWGG sites, resulting in 31 nt CCMD reads (dotted vertical line). All numbers are for reference-matched reads. (A) Total number of reads of each length. (B) Mean read copy number of each length. (C) Fraction of reads of each length that passed base filtering.

Fig 5. Bar graph of random read analysis.

For each combination of G+C content and fraction of non-CCMD reads (horizontal axes), we determined the largest number of reads at which the motif was inaccurately called and added one to this value. The number of reads required to accurately call the motif (vertical axis) was calculated as the mean of 25 replicate determinations.