SAN FRANCISCO (GenomeWeb) – Researchers from the Garvan Institute of Medical Research in Sydney have evaluated methods for analyzing the epigenetic mark 5-hydroxymethylcytosine (5-hmC).
The researchers reported this week in the journal Epigenetics & Chromatin that although whole-genome approaches with single-base resolution provide the most quantitative results, they can be costly, while an array-based method is cost effective and simple to perform, but analysis is restricted to specific locations.
Clare Stirzaker, a co-lead author of the paper and leader of the Epigenetic Deregulation in Cancer group at Garvan, said that the 5-hmC epigenetic mark was only recently discovered in 2009, so its role is not well characterized. Her group is interested in trying to understand the function of 5-hmC and its role in modulating DNA methylation and demethylation.
"We really wanted to take a genome-wide view," she said, but did not know what the best approach would be. The gold standard for evaluating a more commonly studied epigenetic mark, 5-methylcytosine, is bisulfite sequencing, but that approach actually cannot distinguish between 5-mC and 5-hmC, she said.
So, the researchers set about identifying available methods and evaluating them. Although whole-genome bisulfite sequencing (WG BS-seq) alone cannot distinguish between 5-mC and 5-hmC, it can be combined with oxidative bisulfite sequencing (OxBS-seq) to distinguish between the two. OxBS-seq, which was developed by researchers at the University of Cambridge in 2012, involves a two-step conversion that turns hydroxymethylated and unmodified versions of cytosine into uracil, enabling 5-mC to be measured directly. By also performing WG BS-seq, which enables the detection of both 5-mC and 5-hmC, the 5-hmC pattern can be inferred.
Since bisulfite sequencing is considered the gold standard, Stirzaker said the team wanted to include that in their evaluation. In addition, an antibody-based approach known as hMeDIP-seq is "widely used," Stirzaker said, so the team included that.
More recently, an array-based approach that uses Illumina's Infinium Human Methylation 450K BeadChip in combination with oxidative bisulfite conversion has become available, she said. And finally, the group used loci-specific TET-assisted bisulfite sequencing to validate candidate regions.
In the study, they tested the three approaches on brain DNA, because the levels of 5-hmC have been found to be the highest in brain tissue, where it is estimated to constitute between .15 percent and .6 percent of the total nucleotides, but is an order of magnitude less abundant in other tissues. And, in human cell lines it is present at even lower levels, approximately .009 percent of total nucleotides, according to the study. But, they also looked at the abilities of the methods to evaluate 5-hmC in a cancer cell line.
The researchers found that the three different approaches had high levels of overall correlation, but each had slight advantages, depending on the specific question being asked, Stirzaker said.
The main advantage of the BS-seq and OxBS-seq combination is that it is quantitative and enables single-nucleotide resolution, Stirzaker said. But, she said, the array-based approach was able to identify regions of lower levels of 5hmC more readily, she said. The BS/OxBS-seq approach needed deeper sequencing to detect lower levels of hydroxymethylation. The hMeDIP-seq approach is semi-quantitative, she said. "Any enrichment-based approach will not be as quantitative as a single-nucleotide approach," she said.
The array-based approach tended to underestimate the 5-hmC signal, due to CpG islands with "extreme levels of total methylation," the authors wrote. This could be overcome with better normalization approaches.
In prostate cancer cell lines, neither the BS/OxBS-seq or array-based methods could detect 5-hmC. The hMeDIP-seq technique did detect low levels of 5-hmC, but on further investigation, those signals appeared to be false positives.
When DNA has a high abundance of 5-hmC, hMeDIP-seq has "high specificity," the authors wrote. But, at levels below around 4 percent, "hMeDIP-seq signals cannot be used as the representative of the actual 5hmC distribution without further validation."
Determining which method is the most appropriate is a "balancing act," Stirzaker. Researchers have to take into consideration cost, what tissue type or cell line is being evaluated, and available bioinformatics expertise. The array-based approach is the least costly, "but you have to be mindful that you won't get all the regions," she said. The BS/OxBS-seq approach will be the most quantitative, but will also be the most expensive, while the hMeDIP-seq method is good, unless evaluating samples where 5-hmC levels are thought to be very low.
Stirzaker said that the researchers did not evaluate methods on the Pacific Biosciences or Oxford Nanopore Technologies platforms, which could enable direct detection of epigenetic marks. At the time, she said that although the team did have access to a PacBio instrument, the methods for epigenetic analysis were just being developed. And, while she said the group has considered evaluating Oxford's MinIon, "we're fairly happy with the approaches we understand now," she said.
She added that the lab is now looking to use these methods to study methylation and hydroxymethylation in normal and diseased tissue, and to try and figure out the role of 5-hmC specifically. "We're interested in understanding the modulation of the methylation landscape," she said. Researchers are now realizing that that regions once thought to be composed of 5-mC may actually be a mixture of 5-mC and 5-hmC, she said, and "we're trying to tease those apart."