Reviewed as one half of a paired batch with ALKBH1. The pairing is not thematic convenience:
a 2025 EMBO J paper proposes that YTHDF3 is the reader of N6-methyladenine (6mA) in genomic
DNA and that it recruits ALKBH1 to erase it. Both genes therefore rest on the same contested
premise — that mammalian genomic 6mA is a real, regulated mark.
YTHDF3 is one of three paralogous cytoplasmic YTH-domain m6A "reader" proteins (YTHDF1/2/3, or
DF1/DF2/DF3). Its YTH domain binds N6-methyladenosine in mRNA; its N-terminal intrinsically
disordered region drives condensation. GOA carries GO:1990247 N6-methyladenosine-containing RNA
reader activity with IBA, IEA and six independent IDA annotations spanning 2012–2020. This
is as well-supported as molecular functions get, and I treat it as the core function.
Note on two of those IDAs whose abstracts foreground a paralog: PMID:24284625 is titled for
YTHDF2, but the cached full text shows GST-YTHDF3 expression constructs and a rabbit anti-YTHDF3
antibody in the methods — i.e. YTHDF3 was directly assayed. This is exactly the situation CLAUDE.md
warns about, and it is a reason to defer to the curator, not to suspect mis-attribution.
GOA carries both sides of this dispute, and it is worth stating plainly.
For translation promotion (2017, two papers published back-to-back in Cell Research):
These underlie GO:0045727 positive regulation of translation (IMP), GO:0045948 positive
regulation of translational initiation (IDA) and GO:0043022 ribosome binding (IDA).
Against (2020, Zaccara & Jaffrey, Cell — the "unified model"):
That paper is the source of the GOA row NOT|involved_in GO:0045948 (IDA) — a negative
assertion of the same term that PMID:28106076 asserts positively.
Position taken. GO:0045948 is marked UNDECIDED on both rows: the positive and the negated
annotation contradict each other, both are IDA, both come from competent labs, and neither can be
adjudicated from the cached record. Recording this as a live disagreement is more informative than
picking a winner. GO:0045727 and GO:0043022 are kept but demoted to non-core, since the
redundancy result means that whatever translation effect exists is not the paralog's defining
activity. The degradation arm (GO:0061157, GO:0043488) is what both camps agree on, so that
is what is marked core.
A note on redundancy and "core": the unified model implies that single-gene phenotypes understate
YTHDF3's contribution, because DF1/2/3 compensate. That is an argument for keeping the reader
activity core even though YTHDF3 knockdown alone is often mild.
Real and mechanistically coherent (multivalent m6A scaffolds + low-complexity domains), but
downstream of the reader activity and stress-conditional, hence non-core. GOA's
GO:0070925 organelle assembly from PMID:31292544 is far too general for what was shown;
the specific granule-assembly terms (GO:0034063, GO:0033962) are the right level.
This supports GO:0003723 RNA binding (IDA), GO:0043488 (IMP) and
GO:1901163 regulation of trophoblast cell migration (IMP).
Caveat worth recording: the premise that mRNA carries abundant m1A had already been substantially
deflated the year before —
The m1A-seq used in PMID:32194978 is the technique that paper challenges. The binding and
degradation experiments themselves are not overturned by this, and none of the annotations is
removed — but a reader annotation whose substrate is transcriptome-wide-rare is weak ground for
"core". This is the third time in this two-gene batch that an antibody-based mapping method
turns out to be the load-bearing element of a substrate assignment (6mA-DIP for DNA 6mA,
m1A-seq for mRNA m1A, and pyridine-borane sequencing for RNA f5C in PMID:41219180).
The 2025 EMBO J paper proposes:
Its own abstract concedes the problem it is trying to solve:
Explicitly recorded finding: this claim has not entered GOA. A live QuickGO query for
UniProtKB:Q7Z739 (2026) returns 17 distinct GO terms — P-body, RNA binding, mRNA binding,
protein binding, cytoplasm, cytosol, cytoplasmic stress granule, cytoplasmic stress granule
assembly, ribosome binding, regulation of mRNA stability, positive regulation of translation,
positive regulation of translational initiation, negative regulation of type I interferon-mediated
signaling pathway, mRNA destabilization, organelle assembly, regulation of trophoblast cell
migration, and N6-methyladenosine-containing RNA reader activity. None of them is a DNA-binding,
chromatin or 6mA term. No GO:0003677 DNA binding, no methylated-DNA-binding term, nothing.
That restraint is correct on the current evidence, and it is worth recording as a positive
observation rather than an omission. The reasons not to annotate it are the same as for ALKBH1:
If genomic 6mA is independently confirmed in mammals, GO:0003677 plus a methylated-DNA-binding
term would become appropriate for YTHDF3. Until then, recording the claim in reason and
suggested_questions — as done here — is the right level of commitment.
GO:0005515 protein binding (six IPI rows: HIV-1 capsid from PMID:22190034, YTHDF1/Q9BYJ9 fromMARK_AS_OVER_ANNOTATED per project guidance. The YTHDF1 interaction is biologically real andGO:0060339 negative regulation of type I interferon-mediated signaling pathway (ISS + IEA fromGO:0003723 RNA binding — a true but generic parent of the informative m6A-reader term thisYTHDF3's core function is what it has always been: a cytoplasmic m6A reader that, redundantly
with YTHDF1 and YTHDF2, binds m6A-modified mRNAs and directs them to destabilisation. The
translation-promoting role is genuinely disputed and GOA carries both sides, including an explicit
NOT annotation. The newly proposed DNA-6mA-reader role has not entered GOA and should not until
the existence of the substrate is settled independently.