TET3 - the CXXC reader whose domain prefers 5-carboxylcytosine, and the only CXXC-family gene ever found mutated in a KMT2A-rearranged cohort
also called TET3 · ten-eleven translocation 3 · TET3FL · TET3s · Cxxc10-1 · Methylcytosine dioxygenase TET3
TET3 carries the same kind of CG-reading module as the KMT2A fusion, but it does not read the same letter: its favourite target is a chemically modified cytosine, not a plain unmodified one. It is also the only member of this whole protein family that has ever turned up mutated in a study of KMT2A-rearranged leukaemia - and that was in older children, not infants.
The same thing, for a scientist
TET3 (O43151, 1,795 aa) carries a CXXC-type zinc finger at **50-90** [E1]. The full-length isoform TET3FL binds unmethylated CpG, but its **highest affinity is toward 5-carboxylcytosine**, and the crystal structure of the CXXC-5caC-DNA complex shows it reading CcaCG sequences [E2]. A direct head-to-head against the KMT2A domain sharpens this: **MLL1-CXXC is specific only for CpG and tolerates no cytosine modification; TET3-CXXC does not require the CpG context and tolerates methyl-, formyl- and carboxyl-modifications preceding G** [E3]. The two domains have similar folds and different jobs, which is the single cleanest demonstration in this family that shared architecture does not mean a shared address. TET3 also has isoforms with and without the CXXC module - in mouse the *Tet3* locus is adjacent to *Cxxc10-1* and alternative transcripts incorporate that CXXC sequence in cis [E4]. **Rule-91 note: UniProt gives TET3 no experimental subcellular localisation at all** - Nucleus, Cytoplasm and Chromosome are each ECO:0000250, by similarity [E1], which is unusual for a protein of this prominence and means every localisation statement about human TET3 in this graph would be an inference. The leukaemia hook is specific and small: **TET3 is one of eleven epigenetic regulatory genes carrying somatic mutations across 9 of 20 (45%) non-infant KMT2A-rearranged leukaemias, against 3 of 22 (14%) infant cases, P = 0.04** - and it is the **only** CXXC-family gene named anywhere in that study [E5]. In DepMap it is nothing: difference +0.007, rank 10,036 of 17,728 [E6].
This is a prediction, not a measurement. No one has seen this protein in this shape. The colours say how much the software trusts itself, and that is worth reading before anything else — the orange stretches are places the model is effectively saying this part has no fixed shape.
I have never read one of these pictures before
A protein is one long chain of amino acids that folds up. The ribbon traces that chain from one end to the other; it is not a wire, it is a simplification of where the backbone runs. Drag to rotate, scroll to zoom.
Domains are the stretches of chain that do a particular job, and they are what the coloured blocks above the picture mark out. Click one and the view flies to it. In this project the important question about any domain is usually whether the fusion still has it — the labels say lost where it does not.
Anything in yellow sticks is a bound small molecule, which is usually the drug.
A CXXC domain that reads a different letter
Jin et al. characterised the full-length Tet3 isoform and found that its N-terminal CXXC domain binds unmethylated CpG but with highest affinity toward 5-carboxylcytosine, solved the CXXC-5caC-DNA co-crystal, and mapped Tet3FL precisely at transcription start sites of lysosome, mRNA-processing and base-excision-repair genes in neuronal cells 2. Their model is that Tet3FL regulates 5caC removal by base-excision repair - a reader of the product of demethylation rather than of its endpoint.
Stroynowska-Czerwinska et al. then ran MLL1-CXXC and TET3-CXXC side by side, controlling for something no previous study in this family had: interference from cytosine bases elsewhere in the substrate. MLL1-CXXC came out specific only for CpG and tolerant of no modification; TET3-CXXC did not require CpG context and tolerated 5mC, 5fC and 5caC preceding G 3. Their proposed explanation is a register shift in the TET3 complex, which makes TET3 binding context-dependent in a way MLL1 binding is not.
The consequence for the beat's question is direct. If TET3 and KMT2A read different chemical states of the same dinucleotide, they are not competitors at the same site - they are sequential occupants of a site whose chemistry is changing. That is a substantively different model from competition and it has a testable prediction: TET3 occupancy should anti-correlate with KMT2A-fusion occupancy across the 5caC gradient, not overlap with it.
The one mutation in the index literature
In the whole-genome and whole-exome study of KMT2A-rearranged leukaemia, somatic mutations were found in eleven epigenetic regulatory genes - CHD4, SETD2, CREBBP, L3MBTL3, ATR, KAT6A, KDM6A, NSD1, PARP8, SUPT3H and TET3 - in 9 of 20 (45%, Wilson 25.8-65.8%) non-infant KMT2A-rearranged leukaemias, compared with 3 of 22 (14%, Wilson 4.7-33.3%) infant cases, P = 0.04 5. TET3 is the only one of the twelve CXXC-family genes named anywhere in that paper - a full-text search returns zero for TET1, TET2, DNMT1, KDM2B, CXXC4 and CXXC5 5.
What can and cannot be said:
- Can: at least one KMT2A-rearranged leukaemia in that series carries a somatic TET3 mutation, and it is in the non-infant arm.
- Cannot: any frequency for TET3 itself. The 9/20 is for the eleven-gene set, not for TET3, and the per-case assignment lives in a supplementary table this beat did not parse. A reader who quotes '45%' for TET3 will be wrong by up to an order of magnitude.
The authors' own interpretation of the infant-versus-older split is worth quoting because it is the beat's question in their words: "the target cell in infants may already have a chromatin state that is more permissive to transformation by the MLL gene rearrangement" 5. That is an argument that the address is a property of the cell, made by the people who did the sequencing, and it is not made through TET2.
What UniProt does and does not know about where TET3 is
TET3's UniProt record carries three subcellular-location terms - Nucleus, Cytoplasm, Chromosome - and all three are ECO:0000250, inferred by similarity 1. There is no ECO:0000269 localisation call and no isoform-scoped one. That is a different failure from TET1's (where the experimental calls exist but are isoform-scoped and the unscoped one is by-similarity) and from the TP53 case Q72 found. It means:
- Any entry in this graph that records 'TET3 is nuclear' as a fact is recording a curator's inference.
- The
check-pi-dimensions.pylocalisation line, which is field-based, would count this entry as covered. It is covered by an inference, and this section is the place that says so.
The functional literature does place Tet3FL at transcription start sites in neuronal cells by genomic mapping 2, which is a chromatin-occupancy measurement and is stronger than the UniProt annotation for the purpose of asking where the protein works - but it is mouse neurons, and it is not a statement about the human protein's compartment.
| uniprot | O43151 |
|---|---|
| gene symbol | TET3 |
| protein name | Methylcytosine dioxygenase TET3 |
| length AA | 1795 |
| alphafold ID | AF-O43151-F1 |
| accession resolution | UniProt REST gene_exact:TET3 AND organism_id:9606 AND reviewed:true, run 2026-09-21; one row; primary gene name read back as TET3 and length 1795 from the same record. |
| domains note | Only the CXXC boundary comes from the UniProt feature table. The catalytic region carries a PF12851 cross-reference but no UniProt Domain feature, so no residue range is asserted for it. |
| cxxc ligand preference | binds unmethylated CpG but with highest affinity toward 5-carboxylcytosine (5caC) in a CcaCG context; does not require CpG context, unlike the KMT2A CXXC |
| classification | CXXC-family reader with RELAXED and shifted specificity - the family's clearest counter-example to the label 'unmethylated-CpG reader' |
| shown for fusion | Somatic mutation present in the non-infant arm of the index KMT2A-rearranged cohort (one of eleven epigenetic regulators, 9/20 cases collectively); no functional experiment in any KMT2A fusion system |
| isoform hazard | Three major isoforms; UniProt names the canonical O43151-1 'isoform 4', which is a numbering trap of its own. Only the full-length CXXC-containing isoform was characterised for 5caC binding. |
| cell type evidence | mouse neuronal cells (occupancy and 5caC binding); mouse ES cells (isoform work); human non-infant KMT2A-rearranged leukaemia exomes (somatic mutation). No haematopoietic functional data. |
| epigenetic readout | 5caC binding; 5mC to 5hmC oxidation; occupancy at transcription start sites |
| subnuclear localization | Not established for the human protein: all three UniProt location terms (Nucleus, Cytoplasm, Chromosome) are ECO:0000250, by similarity, and there is no experimental localisation annotation. Functionally, mouse Tet3FL maps precisely to transcription start sites in neuronal cells (PMID 26774490). |
| cell type | Mouse neuronal cells (occupancy and 5caC binding); mouse ES cells (isoform work); human NON-INFANT KMT2A-rearranged leukaemia exomes, where a somatic TET3 mutation was found among eleven epigenetic regulators. No haematopoietic functional data and nothing in infants. |
domains
[
{
"name": "CXXC (CpG-binding zinc finger)",
"start": 50,
"end": 90,
"pfam": "PF02008"
}
]localization
{
"subcellular": "nucleus, cytoplasm and chromosome - ALL THREE ECO:0000250 (by similarity); UniProt records no experimental localisation for human TET3",
"subnuclear": "Tet3FL maps to transcription start sites by genomic mapping in mouse neuronal cells; no human subnuclear data",
"localization_shown_for_fusion": "NO - nobody has looked in any KMT2A-fusion system"
}essentiality
{
"depmap_scan": "analyses/depmap-kmt2a-selective-scan-full",
"difference_kmt2a_r_minus_other": 0.0069,
"mean_kmt2a_rearranged": -0.0297,
"mean_other_leukemia": -0.0366,
"pan_essentiality_all_lines": -0.0682,
"n_all_lines": 1208,
"rank_of_17728": 10036,
"percentile": 56.611,
"release_note": "Open Targets dataVersion 26.06, read 2026-09-21"
}- A CXXC-directed ligand designed against the unmethylated-CpG state may miss TET3 entirely - which is good for selectivity - or may hit it hard in cells where 5caC is abundant. Selectivity measured on one substrate state is not selectivity.
- TET3 is the one CXXC-family gene with a somatic mutation on record in this disease, and this graph had no entry for it until now. Any resistance or stratification analysis that enumerated epigenetic co-mutations from the graph would have missed it.
- Which case in the non-infant KMT2A-rearranged cohort carries the TET3 mutation, what kind of mutation is it, and is the CXXC domain affected? Supplementary tables 42 and 44 of that paper hold the answer and were not parsed here.
- If TET3-CXXC prefers 5caC and MLL1-CXXC refuses any modification, do their genomic occupancies anti-correlate? A single published ChIP pair in one cell type would answer it and none exists.
- Is 5caC abundant enough at CpG-island promoters in haematopoietic cells for the TET3 preference to have any bearing on the fusion's address, or is it a property of neurons?
- Why does human TET3 have no experimental subcellular localisation in UniProt, and does a primary paper contain one that curation has missed?
- E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database
TET3 (O43151) is 1,795 aa with a CXXC-type zinc finger at 50-90, its curated DOMAIN comment states the CXXC mediates binding to CpG-DNA, and all three of its subcellular-location terms carry only ECO:0000250 (by similarity) with no experimental localisation annotation.
UniProtKB O43151 (TET3_HUMAN), Methylcytosine dioxygenase TET3 · UniProt Consortium · UniProtKB 2026What this evidence cannot show
Fetched 2026-09-21; accession resolved by gene_exact:TET3 AND organism_id:9606 AND reviewed:true, one row, primary gene name read back as TET3 and length 1795 from the same record. The absence of an experimental localisation code is a fact about the annotation, not about the protein; primary papers may contain localisation data UniProt has not curated.
Evidence that argues against this (1)
- Jin et al. map Tet3FL genome-wide in neuronal cells to transcription start sites, which is a direct chromatin-occupancy measurement that the UniProt annotation does not reflect. Jin SG et al. 2016
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
The full-length TET3 CXXC domain binds unmethylated CpG but has its highest affinity toward 5-carboxylcytosine, with a crystal structure of the CXXC-5caC-DNA complex showing it reads CcaCG sequences, and TET3FL maps to transcription start sites of lysosomal, mRNA-processing and base-excision-repair genes.
Tet3 Reads 5-Carboxylcytosine through Its CXXC Domain and Is a Potential Guardian against Neurodegeneration. · Jin SG, Zhang ZM, et al. · Cell Rep 2016What this evidence cannot show
Binding measured on synthetic oligonucleotides with the isolated domain; the genomic mapping is in mouse neuronal cells. Three major isoforms exist and only TET3FL was characterised in this way. Whether 5caC is abundant enough at CpG islands in a haematopoietic cell for this preference to matter is not addressed.
Evidence that argues against this (1)
- Liu et al. report that in vitro Cxxc4 and the isolated CXXC domains of Tet1 and Tet3(CXXC) bind DNA substrates with SIMILAR preference toward the modification state of cytosine at a single CpG site, which conflicts with a distinctive 5caC preference for TET3. Liu N et al. 2013
- E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper
In a direct comparison controlling for interference from cytosines elsewhere in the substrate, MLL1-CXXC is specific only for CpG and tolerates no cytosine modification, whereas TET3-CXXC does not require CpG context and tolerates methyl-, formyl- and carboxyl-modified cytosine preceding G.
Specificity of MLL1 and TET3 CXXC domains towards naturally occurring cytosine modifications. · Stroynowska-Czerwinska A, Piasecka A, et al. · Biochim Biophys Acta Gene Regul Mech 2018What this evidence cannot show
Isolated recombinant domains on synthetic duplexes. The paper's own point is that earlier work in this family did not control for flanking cytosines, which means every earlier specificity number - including some this graph quotes elsewhere - carries that uncontrolled variable. No cellular or chromatin data.
Evidence that argues against this (1)
- Single-molecule work on the MLL1 CXXC domain with a long DNA carrying a real CpG island shows spatially dependent collaboration among CpGs, so a specificity measured one duplex at a time may not describe behaviour on a genomic island. Liang L et al. 2021
- E4moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper
In human and mouse the TET3 gene is adjacent to a CXXC-encoding locus (Cxxc10-1 in mouse), alternative Tet3 transcripts incorporate that CXXC sequence, and Tet3 also interacts with Cxxc4.
Intrinsic and extrinsic connections of Tet3 dioxygenase with CXXC zinc finger modules. · Liu N, Wang M, et al. · PLoS One 2013What this evidence cannot show
Mouse. The alternative transcripts are evidenced at RNA level and the Tet3-Cxxc4 interaction is biochemical; the inference that distinct ratios of Tet3(CXXC) isoform and Tet3-Cxxc4 complex exist in adult tissues comes from relative transcript levels, not from protein measurement.
Evidence that argues against this (1)
- The same paper reports that Tet1 and Tet3 isoforms with and without a CXXC domain hydroxylate genomic 5mC with similar activity in vivo, which weakens the case that the CXXC module is what determines TET targeting. Liu N et al. 2013
- E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
TET3 is one of eleven epigenetic regulatory genes carrying somatic mutations across 9 of 20 non-infant KMT2A-rearranged leukaemias versus 3 of 22 infant cases (P = 0.04), and is the only CXXC-family gene named anywhere in that study.
The landscape of somatic mutations in infant MLL-rearranged acute lymphoblastic leukemias. · Andersson AK, Ma J, et al. · Nat Genet 2015What this evidence cannot show
The 9/20 is the frequency for the eleven-gene SET, not for TET3; no TET3-specific frequency can be derived from the main text, and the per-case assignments are in a supplementary table this beat did not parse. Non-infant arm is 20 whole exomes spanning ALL, AML and one acute undifferentiated leukaemia, so it pools lineages. The 'only CXXC gene named' claim comes from a full-text term search of 71,489 characters and does not cover supplementary tables.
Evidence that argues against this (1)
- The comparison the paper reports is between age groups within KMT2A-rearranged disease and has no KMT2A-wild-type comparator, so 45% in older children cannot be called enrichment relative to leukaemia in general. Andersson AK et al. 2015
- E6moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset
TET3 shows no selective dependency in KMT2A-rearranged cell lines: difference +0.0069, rank 10,036 of 17,728, pan-essentiality -0.0682 across 1,208 lines.
Genome-wide KMT2A-rearranged selectivity scan, 17,728 genes (DepMap CRISPR Chronos via the Open Targets Platform GraphQL API) · Oncofusion knowledge-graph project · analyses/depmap-kmt2a-selective-scan-full/results.json, generated 2026-09-19; Open Targets dataVersion year 26 month 06, re-read from the API on 2026-09-21 2026What this evidence cannot show
The scan compares 11 Cellosaurus-annotated KMT2A-rearranged lines against 43 other leukaemia lines; none of the 11 is infant-derived and the arm is myeloid-enriched, so a difference mixes fusion status with lineage. Both the difference and the pan-essentiality across all 1,208 lines are quoted, because a gene required everywhere is not a selective dependency. Open Targets dataVersion 26.06; DepMap's own release identifier is not returned by that endpoint.
Evidence that argues against this (1)
- TET3 has paralogues TET1 and TET2 that are also near-zero in the same scan, so a single-gene CRISPR screen cannot exclude a buffered family-level requirement. Oncofusion knowledge-graph project 2026
Head-to-head: MLL1-CXXC tolerates no cytosine modification, TET3-CXXC tolerates 5mC, 5fC and 5caC and does not require CpG context - the sharpest published contrast between the index domain and a family member.
A CXXC reader whose preferred ligand is 5-carboxylcytosine, so it occupies the address in a different chemical state rather than competing for the same one.
One of the twelve human CXXC proteins, and the family's clearest specificity outlier.
The only CXXC-family gene found somatically mutated in the index KMT2A-rearranged cohort, in the non-infant arm; that enumeration did not include it.
The two TET paralogues that do carry a CXXC domain; TET2 does not.
2-HG is a competitive inhibitor of the TET family as a class, not of TET2 alone.