DNA cytosine methylation cycle — de novo CpG methylation, UHRF1/DNMT1 maintenance, TET/TDG active demethylation and methyl-CpG readers; DNMT3A/DNMT3B/DNMT3L/UHRF1/DNMT1/TET1-3/TDG/MBD1/MBD2/MECP2/ZBTB33/ZBTB4/ZBTB38/ZFP57

In mammalian genomes most CpG dinucleotides carry 5-methylcytosine (5mC), whereas CpG islands at promoters are usually unmethylated. Methylation patterns are written, copied, erased and read by distinct sets of proteins that use the same nucleotide as their substrate. De novo methylation, which establishes patterns in the germline and the early embryo, is carried out by DNMT3A and DNMT3B. These transfer a methyl group from S-adenosyl-L-methionine to C5 of cytosine in unmethylated DNA. DNMT3L, which has no catalytic activity, binds the DNMT3A catalytic domain, stimulates it, and reads histone H3 tails unmethylated at lysine 4; it is needed to set up maternal imprints. After replication, CpGs are hemimethylated, carrying 5mC only on the parental strand. Maintenance methylation copies the pattern onto the daughter strand. UHRF1 binds hemimethylated CpG through its SRA domain and H3K9me3-marked chromatin through its tandem Tudor domain. Its RING domain then mono-ubiquitinates histone H3 at lysines 18 and 23. The replication foci targeting sequence of DNMT1 binds this dual-ubiquitin mark, which recruits DNMT1, relieves its autoinhibition and lets it methylate the daughter-strand cytosine. Active demethylation begins with the TET dioxygenases (TET1, TET2, TET3). These Fe(II)- and 2-oxoglutarate-dependent enzymes oxidise 5mC in turn to 5-hydroxymethylcytosine (5hmC), 5-formylcytosine (5fC) and 5-carboxylcytosine (5caC). Thymine DNA glycosylase (TDG) excises 5fC and 5caC, but not 5mC or 5hmC, and base excision repair then restores unmodified cytosine. Methylation also passes without repair, because 5hmC is not a substrate for DNMT1 maintenance. Methylated CpG is read by two structural families of proteins, separate from the SRA reader UHRF1. The methyl-CpG-binding-domain proteins are MBD1, MBD2 and MeCP2. The C2H2 zinc-finger proteins are Kaiso (ZBTB33), ZBTB4, ZBTB38 and the KRAB zinc-finger protein ZFP57. These readers recruit corepressor, histone deacetylase and histone methyltransferase complexes to methylated loci, and ZFP57 protects the methylation of imprinting control regions in the early embryo.

Derived from the RGD Pathway Ontology diagram PW:0001340 (DNA modification pathway) through the RGD_PATHWAY_ONTOLOGY project's diagram->module route; GO mapping GO:0006304 DNA modification (closeMatch). RGD curates 28 human genes to PW:0001340, but the diagram draws only Dnmt1, Dnmt3a and Dnmt3b as gene nodes. The other members sit behind four group nodes ("TET proteins", "TET interacting partners", "MBP proteins", "non-methylated CpG readers"). The module models the 5-methylcytosine cycle as four parts: de novo methylation, maintenance methylation, active demethylation and methyl-CpG reading. Fourteen of the 28 RGD genes are modelled (DNMT1, DNMT3A, DNMT3B, TET1, TET2, TET3, UHRF1, MBD1, MBD2, MECP2, ZBTB33, ZBTB4, ZBTB38, ZFP57). Two genes not on the RGD list are added: DNMT3L and TDG. Boundary decisions follow. (1) De novo methylation is a part whose catalytic step is a variant set of DNMT3A and DNMT3B (ONE_OR_MORE, as the two enzymes overlap but differ in targets, e.g. DNMT3B for centromeric minor satellites). DNMT3L is added as an optional stimulatory annoton with enzyme activator activity GO:0008047. This matches its IDA annotation and the core function in the DNMT3L gene review. DNMT3L is needed for maternal imprints but is restricted to germ cells and the early embryo, which is why it is optional. DNMT3C, the rodent-specific de novo enzyme for young retrotransposons, has no human orthologue. It is not modelled here; modules/pirna_mediated_germline_silencing.yaml covers piRNA-directed methylation and names DNMT3C and DNMT3L in that setting. (2) Maintenance methylation follows the human GO-CAM gomodel:6882d2b800000042, the project's best-overlap model. UHRF1 reads hemimethylated DNA and H3K9me2/3 and ubiquitinates H3K18/H3K23. DNMT1 reads H3K18ub/H3K23ub, and that reading positively regulates its methyltransferase activity. Each activity is linked to its GO-CAM activity. The DNMT1 methyltransferase is grounded on GO:0051718 (acting on CpG substrates), the core function in the DNMT1 gene review. That term is a child of the GO-CAM's GO:0003886. The GO-CAM also models SMARCA3/HLTF (Q14527) as a second H3K23 ubiquitin ligase and H3K9me2/3 reader feeding DNMT1. HLTF is not on the RGD list and is not modelled here. Its evidence (PMID:40680746) shows H3K23 ubiquitination and H3K9me3 maintenance in colorectal cancer cells. That paper raises a role in DNMT1 recruitment only in its discussion, as a possibility, so this is recorded as a knowledge gap. The GO-CAM's part_of GO:0141119 assertion for the HLTF activities may need upstream review. Three further human GO-CAMs model the regulation of UHRF1 stability and H3K9me3 binding: PKA phosphorylation (gomodel:6882d2b800000403), USP7 deubiquitination (gomodel:6882d2b800000491) and CDK1 phosphorylation (gomodel:6882d2b800000559). They reuse the same UHRF1 and DNMT1 activities. They are regulation of the maintenance part and are linked only as secondary associations. USP7, PKA and CDK1 are not modelled. (3) Active demethylation is a two-step chain. The first step is TET oxidation, a variant set of TET1, TET2 and TET3 (ONE_OR_MORE). Each TET carries GO:0070579 DNA 5-methylcytosine dioxygenase activity, which is IDA or EXP for all three. The three Rhea reactions are RHEA:52636 (5mC to 5hmC), RHEA:53828 (5hmC to 5fC) and RHEA:53832 (5fC to 5caC). TDG excision is the second step. TDG is not on the RGD list, but RGD's text says that 5fC/5caC "may be substrates for glycosylases and subsequent targeting by the base excision repair (BER) pathway". TDG is that glycosylase (PMID:21817016, PMID:21862836), and the GO definition of GO:0141167 names it. TDG is therefore added, with GO:0019104 DNA N-glycosylase activity (IDA, IMP). GO has no 5fC- or 5caC-specific glycosylase term; this is an ontology gap, recorded here and not as a knowledge gap. Comparator check (human GOA only; mouse and rat were not checked): GO:0141167 is carried only by TDG (TAS). The TETs carry the sibling regulatory term GO:0044029 (TET1 and TET2 IDA), not GO:0141167. Because the GO:0141167 definition begins with "the enzymatic oxidation of the 5-methylcytosine", and the TETs perform that step, GO:0141167 is placed on the TET annotons. The missing TET GO:0141167 annotations are a curation gap and are not recorded as a knowledge gap. The downstream BER steps (AP site incision by APEX1, gap filling by POLB, ligation) belong to the RGD base excision repair subpathway PW:0000129. No base excision repair module exists in modules/ yet, so they are described but not modelled. Modelling them here would turn a general repair pathway into a part of this module. (4) MBD4 is on the RGD list as an MBD-family methyl-binding protein but is not modelled. It is a mismatch-specific glycosylase that removes T or U from G·T/G·U mispairs at CpG, the product of 5mC deamination (PMID:10499592). It has no activity on 5caC (PMID:21817016). Its role is to repair the mutagenic deamination of 5mC, which belongs to the BER subpathway, not the methylation cycle. Its methyl-CpG binding serves that repair role, not chromatin silencing. GO:0141169 (direct 5-methylcytosine excision) is the plant DME/ROS1 route, and MBD4 is not modelled there either. (5) Methyl-CpG reading is a part whose variant set is the two reader families (ONE_OR_MORE), each with its own nested variant set of proteins. MBD family members are MBD1 and MBD2 (GO:0008327 methyl-CpG binding, IDA) and MECP2 (GO:0010385 double-stranded methylated DNA binding, IMP; MECP2 has no experimental GO:0008327). MBD3, which does not bind methylated DNA (PMID:9774669), is not on the RGD list and is excluded. The zinc-finger family members are ZBTB33/Kaiso, ZBTB4 and ZBTB38 (each GO:0008327, IDA) and ZFP57. ZFP57 is kept as a reader because it binds the methylated TGCCGC hexanucleotide in imprinting control regions and recruits KAP1/TRIM28, DNMTs and UHRF1 to protect that methylation (PMID:22055183). Its function is grounded on GO:0010385, which is IDA for mouse Zfp57 in QuickGO (checked 2026-10-26). Human ZFP57 has only GO:0003682 chromatin binding as experimental support. UHRF2 shares the SRA domain but is not modelled. Its experimental annotations are ubiquitin and SUMO ligase activities, and none of the cited literature shows that it substitutes for UHRF1 in recruiting DNMT1. The readers' downstream effectors (SIN3A and histone deacetylase complexes for MeCP2 and MBD2, SETDB1 for MBD1, KAP1/TRIM28 for ZFP57) belong to the histone modification subpathway PW:0001338 and are not modelled; SIN3A is on the RGD list for this reason. (6) Out of scope. Five transcription factors are context or targets, not participants in the cycle: EBF1, SPI1/PU.1, NANOG, PPARG and PRDM14. They regulate, or are regulated by, methylation at specific loci (TET or DNMT recruitment, pluripotency and germline demethylation programmes), and none of them writes, erases or reads 5mC. The CXXC-domain unmethylated-CpG readers (CXXC1/CFP1, KDM2A, KDM2B; also the CXXC domains of KMT2A, MBD1 and TET1) form RGD's "non-methylated CpG readers" node. They are not a part here. Their reading of unmethylated CpG (GO:0045322 unmethylated CpG binding) recruits histone H3K4 methylation by SET1/COMPASS (CXXC1, KMT2A, KMT2C) and H3K36 demethylation (KDM2A/B) to CpG islands. These are histone-modification outputs belonging to PW:0001338, and they do not act on 5mC. OGT (TET-interacting partner; O-GlcNAcylation of TET and of histones) and SIN3A (TET1 and MeCP2 corepressor partner) are regulators or effectors and are left out under the same rule. Applying CLAUDE.md's substrate-versus-participant test, none of these proteins catalyses a step of methylation, oxidation or excision. The chromatin remodeling subpathway PW:0001339 is also out of scope. S-adenosylmethionine supply by the methionine cycle is substrate supply and is out of scope. Molecular functions come from each protein's experimental GO annotations in QuickGO (checked 2026-10-26). Gene reviews exist for DNMT1, DNMT3A, DNMT3B and DNMT3L, and their core functions were used. The other members have no review yet. No GO-CAM in gocams/index.tsv covers the TETs, TDG or any methyl-CpG reader (checked 2026-10-26). PANTHER family ids and PTNs are deliberately not asserted.