Long-read sequencing maps the hidden 5hmC landscape of human neurons

The study, published in Epigenetics & Chromatin, profiled both 5-methylcytosine (5mC) and 5-hydroxymethylcytosine (5hmC) across the entire genome of human neuronal nuclei, revealing that the two marks occupy distinct and functionally meaningful territories within the neuronal genome.

While 5mC is the best-studied epigenetic modification in DNA, known for its roles in genome stability and transcriptional regulation, 5hmC has remained far more enigmatic. It is generated when TET enzymes oxidize 5mC as part of an active demethylation pathway, but it is also particularly enriched in neurons, hinting at functions beyond being a mere intermediate. Levels of 5hmC in gene bodies, enhancer states and transcription start sites have previously been associated with gene expression in a tissue-specific manner, yet the regulatory mechanisms remain poorly understood.

The technical hurdle has been that conventional methods cannot cleanly separate the two marks. Bisulfite-based approaches such as whole-genome bisulfite sequencing, reduced representation bisulfite sequencing and Illumina DNA methylation arrays convert non-methylated cytosines chemically, degrading DNA in the process, and they cannot distinguish true 5mC from 5hmC, instead reporting a combined signal. Separating the two typically requires an additional parallel treatment of a separate DNA aliquot, either oxidative bisulfite or Tet-assisted bisulfite conversion, with 5hmC then calculated indirectly by subtraction. The new study sidestepped these problems by using Oxford Nanopore Technology, which threads long, native DNA fragments through protein pores and detects modifications as changes in electric current, with deep-learning-based calling algorithms assigning base identities and modification states.

The researchers, led by Dennis Klose and senior author Maja Jagodic, applied the method to neuronal DNA isolated from NeuN-positive nuclei sorted from post-mortem human brain tissue using fluorescence-activated nuclei sorting. The samples came from two donors: a person with multiple sclerosis and a non-neurological control. Sequencing was performed on PromethION flow cells, and methylation calling in CpG context distinguished 5mC and 5hmC simultaneously from the same molecules. The team validated their measurements against bisulfite- and oxidative bisulfite-based Illumina 450K array data generated from the same samples, finding overall high consistency, although agreement decreased at CpGs with high estimated 5hmC levels.

The density distributions revealed a striking difference between the two donors. While total combined 5hmC plus 5mC levels were similar, the individual signals differed: the control sample peaked sharply at 0% 5hmC, whereas the multiple sclerosis sample showed a larger shoulder of intermediate 5hmC values, reflected in median 5hmC levels of 30% versus 13.1%. The authors caution that this comparison was intended only to illustrate the potential of the technology to detect differences in genome-wide 5hmC distributions, and that no conclusions about disease-associated epigenetic alterations can be drawn from it.

Across the genome, the two marks followed broadly similar but not identical patterns. Intronic CpGs showed the highest levels of both modifications, with median 5mC of 52% and 5hmC of 34%, followed by 3’UTR, downstream and coding exon sequences, while promoter-like features such as upstream regions and 5’UTRs showed the lowest levels. DNA methylation levels of both marks dipped at exon-intron junctions. Notably, 5hmC was largely absent from CpG islands, which instead persisted at the high end of the 5mC range, and the proportion of distal enhancer elements increased in regions with higher 5hmC levels, a pattern the authors urge readers to interpret carefully given the few sites with very high 5hmC.

To connect the marks to regulatory function, the team stratified genomic regions according to an 18-state chromatin model summarized across healthy brain samples from the International Human Epigenome Consortium. Combinatorial patterns of 5hmC and 5mC cleanly separated brain-derived chromatin states: promoter-associated states and bivalent enhancers showed low levels of both marks, enhancer states showed intermediate-high 5hmC with intermediate-low 5mC, actively transcribed states showed high 5hmC with intermediate-high 5mC, and heterochromatin and zinc-finger repeats showed intermediate-low 5hmC with high 5mC. The combination of the two modifications, in other words, carries information about neuronal chromatin architecture that neither mark alone provides.

Pathway analysis of the 3,838 genes near regions with high 5hmC above 60% pointed squarely at neuronal biology. Enriched canonical pathways included synapse formation, axonal guidance and neurotransmitter signaling, along with intracellular pathways involving Rho GTPase, protein kinase A and circadian rhythm signaling. The top term, synaptogenesis signaling, implicated adhesion molecules of the ephrin, cadherin, neuroligin, neurexin, reelin, synaptotagmin and synapsin families, together with glutamate receptor subunits, calcium channels and downstream signaling proteins. When genes were stratified by chromatin state, pathways overlapping transcribed regions related to protein synthesis, splicing, neurotransmission and axonal guidance, while enhancer and Polycomb-repressed categories yielded neuron-specific terms such as calcium signaling, myelin signaling and neuroactive ligand-receptor interaction.

The analysis extended to transcription factors. Motif enrichment at CpGs with high 5hmC or high 5mC identified 34 significantly enriched transcription factors in the relative analysis, including Gmeb1, ARNT::HIF1A, TCFL5, E2F8, TFDP1, MAX, E2F6, Ahr::Arnt and Arnt, with additional factors significant in absolute analyses. Nearly 70% of the significant factors were enriched at both high-5hmC and high-5mC sites, but several were specific to one mark: E2F8, E2F6 and TFDP1, known repressors of the cell cycle in post-mitotic neurons, were specific to 5hmC, while SP1, SP2, SP4 and several KLF family members were specific to 5mC. Building regulatory networks from these factors and their predicted target genes produced a dense network around ARNT::HIF1A, CLOCK and MAX, with gene ontology terms pointing to axonogenesis, neuron projection development and neuron migration. The authors note that the approach captured not only active but also repressed regulatory programs, since factors such as ZFP57 and KLF15, expressed predominantly in oligodendrocytes and astrocytes according to Human Protein Atlas single-nuclei data, appeared among depleted targets.

The study’s limitations are clearly stated. It rested on only two post-mortem brain samples from different disease categories, limiting the ability to separate disease effects from inter-individual variability, and the enrichments observed within chromatin states, transcription factor binding sites and pathways should not be read as direct evidence of regulatory mechanisms. The orthogonal comparison with array data, based on 171,999 and 14,750 CpG sites after coverage filtering, should be regarded as technical support for genome-wide profiles rather than locus-specific validation. The work also used the now-decommissioned R9 flow cell chemistry, and the authors expect newer chemistries and improved basecalling models to further refine such insights. Bulk NeuN-positive nuclei at a single post-mortem time point also precluded assessment of temporal dynamics in the regulatory networks. Even so, the team has released an interactive tool, the 5hmC_regulator shinyApp, allowing other researchers to rebuild and explore these networks with alternative parameters, positioning the approach as a framework for future studies of 5hmC and 5mC in brain development and neurological disease.

Subject of Research: Whole-genome profiling of 5mC and 5hmC DNA modifications in sorted human neuronal nuclei using Oxford Nanopore sequencing

Article Title: Whole-genome 5-hydroxymethylation profiling of human neurons for regulatory and biological insight

Article References: Klose, D., Sepehri, M. H., Olsen, R.-A., Vu, H., Ernst, J., Kular, L., Needhamsen, M., & Jagodic, M. (2026). Whole-genome 5-hydroxymethylation profiling of human neurons for regulatory and biological insight. Epigenetics & Chromatin, 19(1), Article 41. https://doi.org/10.1186/s13072-026-00703-z

Image Credits: AI Generated

DOI: 10.1186/s13072-026-00703-z

Keywords: 5-hydroxymethylcytosine, 5-methylcytosine, DNA methylation, Oxford Nanopore sequencing, human neurons, epigenetics, chromatin states, enhancers, transcription factors, post-mortem brain, gene regulation, neuroepigenetics