Taken together, our data indicate a relationship between DNA hydroxymethylation and enhancer activation during mouse cell differentiation. Our observations are in line with previous studies reporting an enrichment in 5hmC at enhancers in correlation with H3K4me1, H3K4me2 and H3K27ac levels in undifferentiated human stem cells (30
) and that binding sites engaged with the pluripotency transcription factor NANOG bear 5hmC. On differentiation, ES cells tend to lose 5hmC, and it remains unknown whether the mark is lost in a general way or if specific genomic regions can become hydroxymethylated during programming of alternative cell states. In this study, we provide evidence that differentiation is associated with dynamic DNA hydroxymethylation of a subset of cell-type-specific enhancers.
Pioneer studies focusing on the genome-wide distribution of 5hmC showed that intragenic 5hmC is positively associated with gene activity (26
). Consistent with this observation, 5hmC inhibits binding of several Methyl CpG binding protein (MBD) including MeCP2 (59
) and could reduce maintenance methylation by DNMT1 (61
). As MeCP2 recruits enzymatic systems that drive chromatin condensation (62–64
), oxidation of 5mC could help chromatin opening and thereby facilitate gene expression. As demonstrated for FOXA1-dependent enhancers, chromatin opening at enhancers is associated with their cell-specific activity (36
). In this study, we establish that chromatin opening, as assessed by FAIRE, is associated with acquisition of 5hmC at selected enhancers. In parallel to the gain in 5hmC, H3K27 acetylation levels raised, suggesting a model of enhancer activation that involves the conversion of 5mC to 5hmC. In line with this hypothesis, a recent study of glucocorticoid receptor (GR)-bound enhancers suggested that a fraction of GR sites distal to TSS and outside of CGIs are demethylated on GR binding (65
). Similar to what is observed for 5hmC-up regions defined in this study, these enhancers are characterized by a peak of CpG density that is also observed for DNAse I hypersensitive sites not overlapping with CGIs (65
), hence defining a category of enhancers that show intermediate CpG density (compared with CGIs and the rest of the genome) and whose activity is controlled by methylation. Recently, Schübeler and coworkers (66
) described methylome remodeling through transcription factor binding at enhancers with low to intermediate levels of CpGs and associated with Tet1. It is most likely that 5hmC-up regions identified in P19 and 3T3-L1 cells belong to this particular category of enhancers implying that loss of methylation at these sites actually results from hydroxylation by Tet enzymes, then leading to chromatin activation. Although we cannot exclude that the observed CpG density of these 5hmC-up regions reflects a bias due to the antibody-based 5hmC capture, they show striking similarities with the aforementioned GR-bound enhancers. Indeed, these regions have intermediate CpG density levels, bind transcription factors and harbor active chromatin marks when hydroxymethylated. We propose that Meis1 activation of enhancers enriched in CpGs involves engagement of Tet enzymes (Supplementary Figure S11
). Indeed, we have shown in this study that Tet1 was recruited to a subset of Meis1-bound enhancers.
The data presented herein argue for a positive role of active DNA demethylation through conversion of 5mC to 5hmC in the regulation of enhancer activity. This could be a transitory state toward replacement of modified cytosines by cytosines through base excision repair (20
), but 5hmC levels seem to be quite stable over time both in P19 and 3T3-L1 cells. Indeed, we observed that enhancers that gain 5hmC at 48
h post RA or DC addition are still hydroxymethylated 7 days after beginning of the treatment (data not shown). This relative stability of 5hmC levels at enhancers suggests that 5hmC could be a signaling mark rather than a short-lived intermediate of active DNA demethylation. Such signaling would involve putative hydroxymethyl-DNA-binding proteins regulating chromatin activity. Alternatively, 5hmC stability could merely protect active enhancer DNA from remethylation and recognition by MBDs. Consistent with the hypothesis that 5hmC is a mark of active enhancer rather than an intermediate in DNA demethylation, we noticed that loss of 5hmC in 5hmC-down regions from P19 cells was correlated with a decrease in active chromatin marks (Supplementary Figure S12
and Supplementary Table S6
). In accordance with chromatin inactivation, motif search in 5hmC-down regions revealed a highly significant enrichment in putative binding sites for pluripotency transcription factors SF1, ESRRB and LRH1 [Supplementary Figure S12d
)]. Importantly, SF1 is known to activate the Oct3/4
promoter in P19 cells, and its expression is dramatically reduced on RA addition (67
). Hence, loss of transcription factor binding during differentiation is likely to promote chromatin inactivation and 5hmC erasure at target enhancers.
Collectively, our data support and extend the concept of transcription factor shaping the methylome at distal regulatory regions by indicating that stable hydroxymethylation of DNA associates with transcription factor binding and activation of enhancers.