DNA cytosine methylation (5mC) is a highly abundant heritable epigenetic mark frequently associated with transcriptional repression, the genomic location of which is central to the processes it regulates, including development, X chromosome inactivation, and human cancers (1,2). Genomic 5mC is commonly mapped using differential enzymatic digestion, bisulfite sequencing, or a combination of both methods (3). The recent discovery of hydroxylmethyl cytosine (hmC) in Purkinje neurons and embryonic stem cells (4,5) via thin-layer chromatography has made it a priority to map this mark on a genome-wide scale to understand its compartmentalization, tissue specificity, and function. Considering the similar structures of hmC and 5mC, it remains unclear how hmC behaves in the classical assays used to measure 5mC. Critically, the presence of hmC in DNA can inhibit the binding of the methyl-CpG binding protein MeCP2 and the enzymatic function of the maintenance methyltransferase DNMT1 (6,7). To test whether restriction digestion can discriminate between 5mC and hmC, we developed an in vitro assay (Supplementary Figure S1) based on PCR amplification, which generates DNA templates that are either unmethylated (dCTP), methylated (dmCTP), or hydroxymethylated (dhmCTP) at cytosine.

We analyzed the human BRCA1 CpG island promoter since it contains numerous methyl-sensitive restriction sites and is known to be hypermethylated in cancer (8). BRCA1 PCR products were digested with the methylcytosine-sensitive enzyme HpaII or its methyl-insensitive isoschizomer MspI (Figure 1A and Supplementary Figure S2). As predicted, digestion of the unmethylated BRCA1 amplicon is complete for both HpaII and MspI, while methylated BRCA1 is resistant to MspI digestion due to PCR incorporation of modified cytosine at the external nucleotide of its recognition site (hmCCGG) (9). Crucially, HpaII digestion of hydroxymethylated BRCA1 is completely inhibited to the same degree as methylated BRCA1, as evidenced by a 60-fold excess of enzyme (Figure 1A). We extended this analysis using the methyl-sensitive enzymes HpyCH4IV, HhaI, and HaeIII and found that digestion of unmethylated BRCA1 is complete for all three enzymes (Figure 1B). In contrast, hydroxymethylated BRCA1 is completely resistant to digestion, indicating a generality in the refractory nature of hydroxymethylated DNA to digestion by methylcytosine-sensitive restriction enzymes. A hydroxymethylated CDH1 (E-cadherin) substrate is also resistant to HpyCH4IV, HhaI, and HaeIII digestion, indicating that hmC inhibition of these enzymes is not sequence-specific (Figure S3). These results suggest that existing vertebrate DNA methylation data generated using methylcytosine-sensitive enzymes may have to take into account the potential presence of hmC in DNA.

Figure 1. Hydroxymethylcytosine-rich BRCA1 DNA sequences are resistant to digestion by methylcytosine-sensitive restriction enzymes. (Click to enlarge)

In vitro sodium metabisulfite (Na₂S₂O₅) treatment of both free and native hmC nucleotides generate the intermediate 5′-methylenesulfonate, which is resistant to deamination (similar kinetics to 5mC deamination) (10). This is in contrast to cytosine, which is readily deaminated. One prediction from this study was that both hydroxymethylated and methylated genomic sequences may be protected from deamination in bisulfite DNA sequencing reactions, suggesting that this technique cannot distinguish these marks. To test this possibility, we bisulfite-treated PCR templates based on the mouse Oct4 (Pou5F1) promoter (prepared with dCTP, dmCTP, or dhmCTP) followed by sequence analysis of cloned products (Figure 2). As anticipated, all cytosine bases in unmethylated Oct4 are successfully converted to uracil bases, which are interpreted as thymine by Taq DNA polymerase. In contrast, cytosine bases in the 5mC-methylated Oct4 template remain unconverted subsequent to bisulfite treatment (Figure 2A). Significantly, hydroxymethylated Oct4 DNA is also completely resistant to chemical modification and is indistinguishable from methylated Oct4 after bisulfite conversion (Figure 2A).

Figure 2. Hydroxymethylcytosine and methylcytosine are indistinguishable subsequent to bisulfite conversion. (Click to enlarge)

Due to PCR incorporation, all cytosine bases in the Oct4 amplicons will be unmethylated, methylated, or hydroxymethylated, including those in non-CpG contexts. To selectively incorporate cytosine in CpG contexts, we used a plasmid clone containing a bisulfite-treated mouse Tex19.1 promoter sequence that contains cytosine bases in the context of CpG dinucleotides only. This approach allowed us to determine if local non-CpG modification of cytosines in Oct4 PCR products impairs bisulfite conversion. Notably, as observed for the Oct4 template, the hydroxymethylated and methylated Tex19.1-derived sequences are completely protected from conversion to uracil/thymine subsequent to bisulfite incubation (Figure 2B). Therefore, non–CpG-modified cytosines do not compromise bisulfite conversion reactions. Together these results indicate that methyl-sensitive enzymes (used in Southern blotting, methylsensitive PCR, COBRA analysis, etc.) and the gold-standard bisulfite sequencing technique (locus-specific or whole-genome bisulfite analysis) are unable to account for the presence of cytosine hydroxymethylation in the genome.

Non-CpG cytosine methylation in human embryonic stem cells has been reported previously using bisulfite sequencing (11,12); from our data we infer that non-CpG hydroxymethylation may exist in these cells and perhaps other tissue types. Moreover, 98% of non-CpG cytosine methylated sites in human ES cell DNA are methylated on one strand only (10). Experiments using the methylcytosine-sensitive enzyme PstI (cuts CTGCAG but not MTGCAG; M = methylcytosine) with CDH1 indicate that hydroxymethylation and hemi-hydroxymethylation in non-CpG contexts inhibit this enzyme activity (Figure S4). Our enzymatic analyses show that many methylcytosine-sensitive enzymes cannot discriminate between methylated, hydroxymethylated, or—in one case—hemi-hydroxymethylated DNA. Furthermore, it is not possible to use bisulfite sequencing to determine if a particular unconverted cytosine base (CpG and non-CpG contexts) in post–bisulfite-treated DNA represents hydroxymethylated cytosine, methylated cytosine, or even incomplete chemical conversion.

In the future, 5mC and hmC discrimination may involve selective immunoprecipitation (with α-5mC) of 5mC-enriched sequences followed by bisulfate analysis of unbound fractions that may be enriched for hmC loci. The reciprocal experiment may be possible when α-hmC becomes available to enrich for bound hmC-enriched sequences. Such strategies will be reliant on α-5mC and α-hmC antibody specificities. To test α-5mC specificity, we probed DNA dot blots containing methylated, hydroxymethylated, and unmethylated BRCA1 and CDH1 DNA, demonstrating that α-5mC (0.1 μg/mL) strongly detects 5mC compared with no affinity for hmC and unmodified cytosine (Figure 2C). It is noted that at higher dilutions (1 μg/mL), α-5mC can partially cross react with hmC (Supplementary Figure S5). To put this in context, the widely used methylated DNA immunoprecipitation (MeDIP) protocol suggests using the identical antibody at 20 μg/mL (13). The difference in 5mC signal between the BRCA1 and CDH1 templates may reflect the relative cytosine content of the amplicons. An additional caveat to the immunoprecipitation approaches is that 5mC and hmC may overlap at common loci, which could hamper selective enrichments.

In summary, current technologies used to assay cytosine DNA methylation do not account for hmC and now necessitate careful re-evaluation of existing methylation datasets. The discovery of hmC will require the development of novel approaches to detect this novel epigenetic mark.