concept · deep · partly written
How DNA methylation is copied through replication
After DNA replication each daughter duplex is half-methylated, and a reader (UHRF1) and a writer (DNMT1) re-methylate the new strand, mostly within minutes but with a measurable failure rate.
Also called: maintenance methylation, DNMT1 maintenance, UHRF1 DNMT1 hand-off
The half-methylated template
A CpG pairs with a CpG on the other strand, so methylation is symmetric. Replication leaves each daughter with one methylated old strand and one blank new strand, a hemimethylated site. That half-pattern is the template the copy machinery reads. 12
Reported, not yet replicated
The reader: UHRF1 more
The SRA domain of UHRF1 binds hemimethylated CpG and flips the methylcytosine fully out of the helix into a pocket, in a mouse structure. Two loops enter from opposite grooves, and the buried interface is about 70 percent larger than that of MBD1, which does not flip. The authors report fully methylated DNA binding about sevenfold more weakly, a figure they take from earlier work, and propose without demonstrating it that a methyl on the second cytosine would clash with Asn494. 1
Reported, not yet replicated
The hand-off is a ubiquitin tag more
UHRF1 is also an E3 ubiquitin ligase. It places two ubiquitins on PAF15, a PCNA-associated factor, at K15 and K24 in humans, or on histone H3 at K14, K18 and K23. The RFTS domain of DNMT1 binds doubly ubiquitylated PAF15 with a dissociation constant of about 1.4 nM; ubiquitin on K15 alone binds far more weakly (about 1.2 micromolar), and 1:1 binding was lost for either single mark. PAF15 is the preferred route in early S phase, histone H3 takes over in late S, and H3 partly compensates when PAF15 is lost. DNMT1 is held inactive by its own RFTS domain; binding of ubiquitylated histone H3 (K18/K23) opens the RFTS lobes in a crystal structure and raised DNMT1 activity in vitro. 23
Reported, not yet replicated
Two speeds: fast and slow phases more
A review describes a fast, replication-coupled phase (minutes, through PCNA and UHRF1-LIG1) and a slower, replication-uncoupled phase (hours, in which UHRF1 reads H3K9 methylation). Hammer-seq measured this at single-molecule level in HeLa and mouse ES cells: the replication-coupled kinetics depend on the UHRF1-Ligase 1 and PCNA-DNMT1 interactions, whereas nucleosome occupancy and UHRF1 binding to methylated H3K9 regulate the uncoupled phase. The uncoupled route was robust enough to largely restore the methylome when the coupled route was severely impaired.
Neighbouring methylated CpGs speed both phases, and isolated CpGs flanked by A or T (solo-WCGW) are slowest. These are also the sites that lose methylation with age and in cancer, and the authors conclude that although quite robust, mitotic inheritance of methylation is imperfect. 45
Reported, not yet replicated
What each fidelity figure is more
The figures in the numbers list are different kinds of quantity and should not be read as one scale. 0.96 and 0.976 are one dataset: Laird reported the first from hairpin-bisulfite data at one CpG island, and Fu refit the same data with a conversion-error term to get the second, so the two are one measurement analysed twice. The Hammer-seq values (over 50 percent at about 4 minutes, over 80 percent at 30 minutes) are a lag after the fork. The remethylation half-lives are within-cycle kinetics from a model. The 88.5 percent value is a snapshot of how many dyads are symmetric.
Estimates from other settings (for example tumour-derived epimutation rates, or gains at unmethylated islands) were seen in secondary reports only and are left out because method and cell identity differ. 67589
Reported, not yet replicated
Where it breaks more
Isolated CpGs lose methylation first. In degron colorectal lines, four days of UHRF1 depletion lowered methylation by at least 25 percent over about 1.3 million 1-kb tiles, versus about 600,000 for DNMT1 depletion, and the difference was not explained by the number of divisions; UHRF1 also affects DNMT3A and DNMT3B and restrains TET2, so 'UHRF1 only hands over to DNMT1' is too simple.
How the numbers mislead: hairpin-bisulfite converts hairpin DNA less efficiently, so Laird kept only sequences with at least 99 percent non-CpG conversion; the figure rests on one locus; Fu's model assumes independent sites while the kinetic reanalysis finds neighbouring rates correlated. 10678
Reported, not yet replicated
The histone side, briefly deep
Histones are copied differently: parental H3-H4 is recycled to both daughters and new histones arrive blank, so a mark is diluted about twofold and rewritten. In ChOR-seq the positions of H3K4me3, H3K36me3, H3K79me3 and H3K27me3 were reproduced on new DNA while levels were diluted; about 50 percent of H3K4me3 regions recovered within 1 hour, and about 80 percent of H3K27me3 regions were restored slowly, up to 24 hours, mostly after mitosis. In vitro, the PRC2 subunit EED binds H3K27me3 and allosterically activates the enzyme, the basis of a read-write model. UHRF1 reads and writes on histones, so the DNA and histone layers share one reader. See the histone records for more. 11124
Reported, not yet replicated
Plain answer
CpG methylation is symmetric: the partner strand carries a mirror-image mark. Replication gives each daughter one old methylated strand and one blank new strand. UHRF1 recognises that half-methylated site, tags a replication-associated protein or histone H3 with ubiquitin, and the tag recruits and activates DNMT1, which methylates the new strand. The best measured figure, at one CpG island, is about 96 to 98 percent of methylated sites re-made per division. 1267
Reported, not yet replicated
Why it matters
Replication is the one moment every mark is physically disturbed, so it is the test of whether epigenetic memory exists. Memory here is a cycle of dilution and repair: a cell stays stable only while repair outpaces loss, and sites that are repaired slowly or missed are where drift starts. 54
Our inference
Real numbers
Per-division maintenance, FMR1 CpG island (hairpin-bisulfite)
About 0.96 per methylated site per division (range 0.95 to 0.98), from 93 dyads, at one hypermethylated CpG island on the inactive X in human cells; de novo methylation on the same island about 0.17. One locus, one source individual per the report that led here. A bisulfite measurement, so conversion error applies. 6
Reported, not yet replicated
The same data refit with conversion error
0.976 (80 percent interval 0.969 to 0.983), with a fitted conversion error of 1.6 percent, over 22 CpGs from 6 females. This is the same dataset as the 0.96 figure refit with an extra error term, not a second independent measurement. The authors note that the rate and the conversion error are partly unidentifiable and that the model assumes independent sites. 7
Reported, not yet replicated
Lag after the fork, HeLa (Hammer-seq)
More than 50 percent of methylated CpGs were maintained within about 4 minutes of fork passage (54 percent in wild type) and over 80 percent by 30 minutes; the remainder took up to about 10 hours. This is a delay after the fork, not a per-division error rate. Abstract read in this run; the figures come from the full text as read by the report checker. 5
Reported, not yet replicated
Post-replication remethylation kinetics, human ES cells
Per-site half-lives from under 10 minutes to 5 hours or more, with rates spanning about two orders of magnitude; fast sites are only bounded from below. A model fit to reanalysed Repli-BS data, within the cell cycle, not a per-division rate. 8
Reported, not yet replicated
Fidelity as a snapshot, mouse ES cells
Fidelity, defined as the share of symmetric CpG dyads, was 88.5 percent as a genome mean, with CpG islands more than 5 points higher. A snapshot of state, not a rate per division. 9
Reported, not yet replicated
Still open here
- Per-division maintenance split by CpG context (isolated versus clustered CpG, island versus non-island) is not published as numbers in the sources opened; Hammer-seq data are public and could be re-analysed, but the accession was not confirmed here.
- No per-division fidelity figure was found for any locus other than one CpG island, in one individual's cells.
- Whether UHRF1 only hands over to DNMT1 is too simple: UHRF1 loss is worse than DNMT1 loss in one cell system, and the reason is not settled.
- The original 1975 proposals of a maintenance enzyme were not opened.
- De novo methylation (DNMT3A and DNMT3B) and the TET-mediated removal path are not covered here.
Not yet written: Equation.
Linked from
- Bisulfite sequencing (WGBS and targeted)
- DNA methylation
- Germline epigenetic reprogramming
- Histone recycling and the PRC2 read-write loop
- Cellular memory
- Can a gene be silenced permanently, and how does cellular memory work?
- Can a histone mark carry memory through cell divisions without a DNA sequence to rewrite it?
- How faithfully is DNA methylation copied at each division, and does it depend on the CpG context?
- How long do X inactivation and imprints last, and how often do they fail?
- Is the irreversibly silent state of an engineered gene a molecular lock, and how long do silenced states last?
Sources
- Hashimoto, H., Horton, J.R., Zhang, X., Bostick, M., Jacobsen, S.E., Cheng, X.. The SRA domain of UHRF1 flips 5-methylcytosine out of the DNA helix. Nature (2008). paper; opened 2026-11-17 all records citing this
- Nishiyama, A., Mulholland, C.B., Bultmann, S., et al.. Two distinct modes of DNMT1 recruitment ensure stable maintenance DNA methylation. Nature Communications (2020). paper; opened 2026-11-17 all records citing this
- Ishiyama, S., Nishiyama, A., Saeki, Y., et al.. Structure of the Dnmt1 reader module complexed with a unique two-mono-ubiquitin mark on histone H3 reveals the basis for DNA methylation maintenance. Molecular Cell (2017). paper; opened 2026-11-17 all records citing this
- Espinosa-Martinez, M., Alcazar-Fabra, M., Landeira, D.. The molecular basis of cell memory in mammals: The epigenetic cycle. Science Advances (2024). review; opened 2026-11-17 all records citing this
- Ming, X., Zhang, Z., Zou, Z., et al.. Kinetics and mechanisms of mitotic inheritance of DNA methylation and their roles in aging-associated methylome deterioration. Cell Research (2020). paper; opened 2026-11-17 all records citing this
- Laird, C.D., Pleasant, N.D., Clark, A.D., et al.. Hairpin-bisulfite PCR: assessing epigenetic methylation patterns on complementary strands of individual DNA molecules. Proc Natl Acad Sci U S A (2004). paper; opened 2026-11-17 all records citing this
- Fu, A.Q., Genereux, D.P., Stoger, R., Laird, C.D., Stephens, M.. Statistical inference of transmission fidelity of DNA methylation patterns over somatic cell divisions in mammals. Annals of Applied Statistics (2010). paper; opened 2026-11-17 all records citing this
- Busto-Moner, L., Morival, J., Ren, H., et al.. Stochastic modeling reveals kinetic heterogeneity in post-replication DNA methylation. PLoS Computational Biology (2020). paper; opened 2026-11-17 all records citing this
- Zhao, L., Sun, M.A., Li, Z., et al.. The dynamics of DNA methylation fidelity during mouse embryonic stem cell self-renewal and differentiation. Genome Research (2014). paper; opened 2026-11-17 all records citing this
- Yamaguchi, K., Chen, X., Rodgers, B., et al.. Non-canonical functions of UHRF1 maintain DNA methylation homeostasis in cancer cells. Nature Communications (2024). paper; opened 2026-11-17 all records citing this
- Reveron-Gomez, N., Gonzalez-Aguilera, C., Stewart-Morgan, K.R., et al.. Accurate recycling of parental histones reproduces the histone modification landscape during DNA replication. Molecular Cell (2018). paper; opened 2026-11-17 all records citing this
- Margueron, R., Justin, N., Ohno, K., et al.. Role of the polycomb protein EED in the propagation of repressive histone marks. Nature (2009). paper; opened 2026-11-17 all records citing this
Where it sits on the map
History of this record (3)
This record has been changed 3 times, on the dates below. The note kept against each change is written for the people who maintain the atlas, in their own shorthand, so it is not shown here; it is in the record itself.
- Status
- partly written
- Last updated
- 2026-11-18
- Record
- content/concepts/dna-methylation-maintenance.json