The human CXXC/CpG-reading family is exactly twelve proteins, and TET2 is not one of them

Three query forms were run on 2026-09-21 and cross-checked against each other, because rule 93 in this project's brief exists precisely because a single query form can be blind to half a database:

The twelve symbols are identical across all three: CXXC1, CXXC4, CXXC5, DNMT1, FBXL19, KDM2A, KDM2B, KMT2A, KMT2B, MBD1, TET1, TET3 1. Dropping reviewed:true returns 158 records whose gene names add only FBXL10 (a retired symbol for KDM2B) and blanks, so the unreviewed tier contributes no new gene 1.

This project's own target-phase2-cxxc-dna-dossier already named the family as the selectivity gate for any CXXC-directed programme, and listed it as "CXXC1/CFP1, MBD1, KDM2A, KDM2B, TET1, TET3, DNMT1, CXXC4/IDAX, CXXC5, KMT2B and others". That list is correct and complete except that there are no "others" - the family is closed at twelve 1.

TET2 is the member of this pathway most often mutated in myeloid malignancy, and it is the obvious candidate for a co-mutation that would change the methylation landscape the KMT2A CXXC domain reads. It is not a CXXC protein. Ko et al. showed that IDAX/CXXC4 was originally encoded within an ancestral TET2 gene that underwent a chromosomal gene inversion during evolution, separating the TET2 CXXC domain from the catalytic domain; the IDAX CXXC domain binds unmethylated CpG, localises to promoters and CpG islands, and interacts directly with the TET2 catalytic domain 2. Liu et al. independently note that in human and mouse the TET2 and TET3 genes are adjacent to CXXC4 and Cxxc10-1 respectively, and that these CXXC domains form a distinct homology group with those of TET1 and CXXC5 5.

So TET2 does reach CpG islands - in trans, through a different gene product. Writing TET2 as a CXXC reader would be a category error; writing it out of the story entirely would miss that the family contains TET2's address module as a separate protein.

Xu et al. measured the DNA-binding selectivities of all known human CXXC domains by several assays and solved a series of CXXC-DNA co-crystals, and the result is a classification by DNA selectivity that cuts across the functional classification - single CXXC domains and tandem CXXC-PHD modules recognise distinct DNA ligands 3. Two worked consequences relevant here:

• TET3's CXXC binds 5-carboxylcytosine with higher affinity than unmethylated CpG, so calling the whole family "unmethylated-CpG readers" is wrong for at least one member [see gene-tet3].
• A direct comparison of MLL1-CXXC and TET3-CXXC finds MLL1-CXXC specific only for CpG and tolerant of no cytosine modification, while TET3-CXXC does not require CpG context and tolerates methyl-, formyl- and carboxyl-modifications preceding G 6.

And the functional test in the disease backbone has been done: substituting the MLL CXXC domain in MLL-AF9 with the DNMT1 CXXC gives robust colony formation and leukaemogenesis comparable to MLL-AF9, while the CGBP/CXXC1 and MBD1 CXXC domains abrogate both - and all substituted fusions still targeted Hoxa9 4. Shared family membership predicts neither shared chemistry nor shared function.

The family is 12 of ~20,000 human protein-coding genes. Eleven of the twelve are nuclear at least in part; CXXC4 is annotated cytoplasmic only, and by similarity (ECO:0000250) rather than experimentally, and FBXL19's experimental UniProt record is cytoplasmic three times over and nuclear once 1. Any claim that the twelve compete for one address has to survive the fact that at least two of them are not reliably in the nucleus.

Coverage in this graph before this beat: 2 of 12 written (gene-cxxc1, gene-kdm2b) plus gene-kmt2a, i.e. 3/12 = 25.0% (Wilson 95% 8.9-53.2%). After this beat, 11/12 = 91.7%; KMT2B remains unwritten and is the one gap, deliberately left because it belongs to the KMT2-family beat that already owns kmt2-family/.