Gene regulatory mechanisms downstream of DNA methylation (paper 17 Sep 26)

#1

https://www.nature.com/articles/s41576-026-01008-3

claude-sonnet-5-paid:

Review: Gene regulatory mechanisms downstream of DNA methylation

Citation: Buckley, Burns, Jarry, Wang, Soe & Jacobsen. Nature Reviews Genetics, published online 17 September 2026 (Review article, 280 references).

1. Summary

This review asks how cytosine methylation (5mC) affects transcription and chromatin once it is in place. The authors focus on “reader” proteins that bind 5mC, and on transcription factors (TFs) that are sensitive to it, in mammals and plants. They organise the paper around three functions.

Maintaining methylation through replication and repair.

  • UHRF1 in mammals and VIM1/2/3 in Arabidopsis recognise hemi-methylated CpG through their SRA domains. They ubiquitinate histone H3 and PAF15, which recruits and activates DNMT1/MET1.
  • CDCA7 works with the remodeller HELLS (DDM1 in plants) to open compact heterochromatin so it can be re-methylated.
  • MBD4 repairs G:T mismatches caused by 5mC deamination. MBD1 links methylation to repair and to replication-fork protection.
  • The authors describe the repair role as apparently mammal-specific, because plants lack MBD4 and MBD1 homologues.

Silencing.

  • In mammals, MBD1 recruits HDAC and SUV39H1-HP1. MBD2 recruits the NuRD complex. MeCP2 recruits NCoR/SMRT in neurons and is tuned by linker histone H1.
  • Methylphilic zinc-finger TFs (KAISO, ZBTB4/38, ZFP57, ZNF445) bind methylated motifs and recruit co-repressors such as NCoR, TRIM28 and SETDB1.
  • In plants, SUVH2/9 link methylation to RNA-directed DNA methylation (RdDM) and MORC/SWI-SNF compaction. SUVH4/5/6 couple non-CG methylation to H3K9me2 through a self-reinforcing loop with CMT2/3.
  • AtMBD2 and AtMBD5/6 read mCG. AtMBD5/6 work with ACD15/21, SLN and HSP70 in a proposed higher-order or condensate-like assembly.

Activation.

  • Some methylphilic TFs, such as KLF4 and FOXA2, recruit TET enzymes to demethylate and open enhancers.
  • In plants, SUVH1/3 with DNAJ1/2 activate genes near methylated transposable elements (TEs) and drive ROS1 expression. AtMBD7 recruits IDM1-mediated H3 acetylation to help demethylases access DNA.
  • In pollen vegetative nuclei, AtMBD5/6 and AtMBD7 appear to compete at shared methylated loci.

Central thesis. Silencing is layered and partially redundant. Direct TF exclusion, several readers, histone marks and RdDM overlap at the same loci. This is why losing DNA methylation itself is severe, while losing any single reader usually gives mild or context-specific phenotypes. The paper cites examples: mammalian MBD mutants, and a recent Arabidopsis preprint in which a combinatorial mutant reproduces about 73% of the silencing loss seen in methylation-deficient plants. The closing questions are how redundant pathways are integrated, and whether 5mC cooperates with RNA m6A and 3D genome organisation.

2. Novelty

This is a review, so the novelty lies in synthesis and framing rather than new data. I judged it against the references the paper cites and my own background knowledge, not a systematic literature search.

  • Reader-centric, cross-kingdom framing. Earlier reviews (several are cited, such as Schmitz 2019, Zhang 2018 and Law & Jacobsen 2010) cover methylation broadly or focus on writers and erasers. Putting mammalian and plant readers side by side, with a unifying table (Table 1), is the main organising contribution.
  • Redundancy as the explanatory principle. The paper offers layered redundancy as a general explanation for why reader mutants have mild phenotypes but methylation loss is catastrophic. It also frames the two competing explanations, TF exclusion versus reader-mediated repression, and says how to test them (remove several readers at once).
  • Treating 5mC as an activating mark. The activation section is given equal weight to silencing. It collects methylphilic TFs, TET recruitment, SUVH1/3, AtMBD7 and the pollen antagonism. This corrects the common view of 5mC as purely repressive.
  • Maintenance and repair as reader functions. Grouping UHRF1, CDCA7-HELLS, MBD4 and MBD1 together as reader-dependent genome-stability mechanisms is a useful reframing.
  • Very recent findings. These include the AtMBD5/6-SLN-HSP70 assembly model, the 2025 pollen single-nucleus multi-omic study of MBD5/6 versus MBD7, CDCA7 function in Arabidopsis, the combinatorial-mutant preprint, and systematic TF methylation-sensitivity screens (23% of 542 human TFs lose binding, about a third prefer methylated DNA, about 76% of 327 Arabidopsis TFs are methylphobic).
  • Reader tools for the field. The glossary, the annotated key references and the closing list of open questions add value.

3. Critique

Strengths

  • The paper is clear and well organised, and its figures are informative.
  • It is candid about unknowns, for example the unresolved mechanism of SUVH1/3 activation and whether AtMBD7 recruits demethylases directly.
  • It notes where mammal-plant homology is uncertain, such as VIM versus UHRF1.
  • It separates derepression from true activation in the glossary.

Weaknesses

  1. The central “framework” is narrative, not formal. The abstract promises a framework, but the redundancy argument is text only. No figure or model integrates the layers, and there is no way to estimate the relative contribution of each layer. The authors themselves list two non-mutually-exclusive explanations for mild reader phenotypes and concede that the key experiment (multi-reader removal) is largely undone. The best direct support is a single preprint, which is not peer reviewed.
  2. Evidence quality is not graded. Table 1 and the figures mix in vitro binding, structural data, single-locus examples and genome-wide genetics without marking which is which. MeCP2 illustrates the problem: its occupancy depends on linker histone H1 and chromatin accessibility, not simply on methylation. Readers cannot easily tell which mechanisms are established in vivo.
  3. Speculative models drawn as settled. The AtMBD5/6 condensate model (Fig. 4c) is supported by findings the text itself describes as “consistent with” condensates. Much of this work comes from the senior author’s own laboratory, and the plant sections lean heavily on that lab’s papers. Independent replication and direct biophysical evidence are not discussed.
  4. Mammal-plant comparison is uneven. The plant side is almost entirely Arabidopsis, with occasional rice and tobacco. The text asserts that mammals rely more on TF exclusion and plants on readers, but does not test that idea or discuss why it might hold evolutionarily. Some paired mechanisms (for example SUVH versus MBD readers) are functionally quite different, so side-by-side presentation risks implying equivalence.
  5. Gaps in scope relative to the abstract.
    • The abstract mentions interaction with “parallel epigenetic systems”, but PRC2/H3K27me3 and other non-H3K9 pathways receive little attention.
    • CTCF and cohesin, which are major methylation-sensitive regulators of 3D genome structure, do not appear to be discussed. 3D genome organisation is raised only in the conclusion.
    • Imprinting, X-inactivation, gene-body methylation function and DNMT3A/3B reading of histone marks are largely absent.
    • Fungi and other lineages are mentioned in the introduction but not covered.
  6. Thin on disease and ageing. The abstract promises relevance to development, differentiation and disease. Coverage of disease is limited to Rett syndrome, ICF syndrome, some cancers and the ageing-related hypomethylation of late-replicating regions. There is little discussion of clinical or therapeutic implications.
  7. Speculative closing. The m6A RNA methylation crosstalk in the conclusion is underdeveloped and not supported by evidence in the body of the paper.
  8. Missing methodological guidance. The review argues that the field needs to disentangle redundant pathways. It could have said more about the tools that would do this: acute degron-based depletion to avoid compensation, locus-specific epigenome editing (only one SunTag-DRM2 example is mentioned), and massively parallel reporter assays that separate TF exclusion from reader recruitment.

Overall assessment. This is a useful, current and well-illustrated reference for the reader landscape, and its strongest contribution is the layered-redundancy perspective. Its main limits are that the thesis is more asserted than demonstrated, the evidence is not graded, and some scope choices sit awkwardly with the abstract. It should be read as an expert’s map of the field, not an independent test of the redundancy hypothesis.

Q2: what does the paper say the function of DNA methylation is

The paper treats DNA methylation (5mC) as a conserved epigenetic mark with several functions. It also argues the field has moved from seeing it as a static repressive mark to seeing it as a dynamic mark that reader proteins and transcription factors interpret.

Core functions stated in the abstract and introduction

  • Regulating gene expression. Methylation influences transcriptional programmes needed for development and differentiation.
  • Repressing transposable elements (TEs). In both mammals and plants this is described as a conserved feature.
  • Maintaining genome stability. The paper also links this to proper development, and says disrupting methylation causes widespread dysregulation, developmental defects, ageing-related change and cancer.

How it acts on transcription (three non-mutually-exclusive mechanisms)

  1. Recruiting readers. Methyl-binding proteins (MBD and SRA domain proteins) bind 5mC and recruit histone modifiers, chromatin remodellers and co-repressors to reinforce transcriptional states.
  2. Shaping chromatin. Reader engagement can stabilise nucleosomes and shift chromatin between closed and open states.
  3. Controlling transcription factor (TF) binding. Methylation blocks many TFs (methylphobic) and attracts some (methylphilic). The paper says direct inhibition of TF binding accounts for much of the repression seen at methylated regulatory regions in mammals, and is the dominant effect in plants.

Specific roles the review describes

  • Silencing. In mammals, readers such as MBD1, MBD2 and MeCP2, and zinc-finger TFs such as KAISO, silence genes and TEs. In plants, RdDM, SUVH proteins and AtMBDs silence TEs, and also silence some genes.
  • Maintaining itself. Readers like UHRF1 (VIM1/2/3 in plants) and CDCA7 restore methylation after DNA replication, so the mark is inherited through cell divisions.
  • Protecting the genome. MBD4 repairs G:T mismatches from 5mC deamination, and MBD1 contributes to repair and replication-fork protection. The paper says these repair links appear to be mammal-specific.
  • Activation in specific contexts. Some methylphilic TFs (KLF4, FOXA2) recruit TET enzymes to demethylate and open enhancers. In plants, SUVH1/3 and AtMBD7 can promote expression of genes near methylated TEs.
  • Overall robustness. The authors argue methylation-dependent repression is rarely a single-step process. It works through overlapping, partly redundant layers, which is why losing DNA methylation itself is severe while losing single readers often gives mild phenotypes.

Differences the paper notes between mammals and plants

  • Mammals: mainly CG methylation, with CpG islands protected from it. Gene-body methylation is variable, and non-CG methylation is mostly confined to neurons and embryonic stem cells.
  • Plants: methylation occurs in CG, CHG and CHH contexts, concentrated on TEs and repeats, and reinforces silencing there. Canonical CpG islands do not exist.

The paper also says some functions remain unresolved, such as how methylation relates to gene activation, and the role of oxidised forms of 5mC in plants.

My own view is that a major role of DNA methylation is to have a mechanism to silence genes so that energy is not wasted on trying to transcribe genes which are neither required not for which the energy is avialable. In particular I think it operates through the gene going dormant and then getting gradually methylated rather than any decision to actually methylate the gene.