Skip to main content
Premium Trial:

Request an Annual Quote

Mayo Clinic Team Shares Comparison of Sequencing, Array Methods for Methylation Profiling


NEW YORK (GenomeWeb) – In an effort to evaluate the similarities and differences between available methods for DNA methylation profiling, researchers from the Mayo Clinic have performed and published a new comprehensive comparison of the approaches, finding that despite differences in coverage, all the methods showed high concordance.

In the study, which appeared online ahead of print last month in the journal Epigenomics, a team led by Mayo Clinic researcher Julie Cunningham set out to compare as directly as possible a set of four sequencing-based and one microarray method for DNA methylation profiling.

"Investigators [now] have a choice of approaches they could use … What we really wanted, was to see if any one of these offered any benefits over another," Cunningham told GenomeWeb this week.

"There have been other comparisons [made], but this was more comprehensive insofar as we added two newer capture methods that have recently come onto the market," she added.

The researchers put these two newer sequencing methods — one called methyl capture sequencing or SS-METH Seq using Agilent capture technology, and another using the Roche NimbleGen SeqCapEpi enrichment platform — up against two other sequencing approaches — reduced representation bisulfite sequencing, and methylated DNA immunoprecipitation, or MeDIP sequencing. Finally, they also evaluated the Illumina Human Methylation 450 Bead Chip, and applied the five techniques to a pair of mother/daughter cell lines.

Overall, the authors found that the five approaches were encouragingly concordant, matching up with each other in both pairwise and four-way analyses with high concordance percentages.

At the same time, the comparison also highlighted important areas where the techniques do differ, with implications for others in the field considering one or another of these techniques for their own epigenetic research.

A central challenge for the study, and a highlight of the investigators' report, was finding ways to accurately compare this group of techniques that differ so significantly from one another in their coverage.

For example, MeDIP differs from bisulfite sequencing in that it is not quantitative, and does not provide single-base resolution of CpG methylation. As an antibody-based approach, it pulls out genomic fragments that are partially methylated just as readily and effectively as those which are highly methylated.

Despite this, the researchers were able to apply a novel analysis approach that allowed them to compare enrichment peaks calculated from the MeDIP data with the methylation levels of the methods boasting single-base resolution.

Performing a pairwise analysis of all the methods except MeDIP, the researchers found high concordances for each matchup — from 0.8 up to 0.89 depending on the two methods being directly compared.

When they looked further, discounting any completely methylated or completely unmethylated CpG's from the comparison, the pairwise concordances were lower, but still remained greater than 0.8, except in the case of RRBS versus SS-MethSeq and RRBS versus the NimbleGen-MethSeq, where the correlation decreased to 0.65.

In a four-way comparison looking at CpGs detected by all the methods, concordance was 79 percent for one cell line, and 77 percent for the other.

When they added in MeDIP using the novel peak estimate approach, the researchers initially saw a positive, but low correlation for MeDIP compared to the methylation ratios of RRBS, SS-MethSeq, or NimbleGen-MethSeq.

However, when the group manipulated its analysis using different methylation value cutoffs, the concordance could be pushed much higher. For example, with a cutoff that dichotomized methylation status into two categories, methylated or not, there was an approximately 98 percent concordance between MeDIP and each of the other sequencing methods.

And even at a cutoff of 0.5, concordance also remained fairly high, between 77 percent and 83 percent, depending on the method being compared.

Finally, the researchers also looked at the different methods' distinction of differentially methylated regions between the mother and daughter cell line samples. Comparing MeDIP with the single-base resolution methods, they found that SS-MethSeq showed the highest number of DMRs in common with MeDIP, followed by NimbleGen capture, and then RRBS.

Because of the lower coverage of the 450K array approach, more than half the sequencing-identified DMRs were not present in the array data. But for those that were, concordance was high compared to both RRBS and the two capture methods — between 84 and 93 percent depending on the method.

"I think we were quite encouraged that the concordance was so high, given that they are such different ways of looking at [methylation,]" Cunningham said. "But we also noted some important differences; for example, that bisulfite-based methods do not distinguish between 5mC and 5hmC — something people should be aware of."

"I think its dependent on what your question is," she added. "If you want to get a really good global overview of methylation, then MeDIP would be the optimal approach … although with the caveat that specific loci cannot be identified in that case."

RRBS, on the other hand, uses the least starting material. "Practically, these choices are so often really driven by cost," Cunningham said. Along those lines, the array approach is most economical. However, as costs come down, NGS methods will become much more competitive, and as sequencing output continues to improve, RRBS would certainly become more of the method of choice, she added, since they do provide so much more information than arrays.

Furthermore, "[RRBS] is certainly the method of choice now if you have limited material, particularly for less-than-ideal DNA, where some of these other methods are not going to be optimal," Cunningham explained.

Cunningham said that at Mayo Clinic, there are numerous investigators looking at DNA methylation, mostly in the context of cancer, including studies comparing tumor and blood samples.