FTO (human, Q9C0B1) — curation journal

Reviewer notes for FTO-ai-review.yaml. Provenance is inline as
[PMID:xxxxx "<verbatim supporting text>"].

1. What the protein is

FTO is a nuclear (and, in some cell lines, partly cytoplasmic) Fe(II)/2-oxoglutarate-dependent
oxygenase of the AlkB superfamily. It was discovered through obesity GWAS and only afterwards
shown to be an enzyme: the founding paper demonstrated 2OG-dependent demethylation of a
methylated nucleobase in single-stranded DNA by the mouse orthologue
PMID:17991826 and its nuclear localization
PMID:17991826.

The fold is AlkB-like with an extra C-terminal domain and a loop that excludes duplex nucleic
acids PMID:20376003. That loop is the structural reason FTO works on single-stranded
substrates only PMID:20376003.

Loss of catalytic activity in humans is not a mild phenotype: the homozygous R316Q variant causes
a lethal recessive polymalformation syndrome
PMID:19559399, establishing that the enzymatic
function matters in vivo PMID:19559399.

2. THE LOCUS IS NOT THE PROTEIN (important for the BP annotations)

The FTO locus carries the strongest common-variant association with BMI, but the causal
mechanism runs through neighbouring genes, not through FTO protein. Claussnitzer et al. showed
the causal SNP disrupts an ARID5B motif in FTO intron 1 and derepresses a preadipocyte enhancer
acting at 1.2 Mb on IRX3/IRX5
PMID:26287746, with the phenotype being a
beige-to-white adipocyte shift
PMID:26287746. Their own conclusion
names IRX3/IRX5, not FTO
PMID:26287746. UniProt records the same caveat verbatim in its OBESITY disease comment
("It is unclear whether variations associated with obesity directly affect FTO function or alter
the expression of adjacent genes such as IRX3, rather than FTO itself").

Curation consequence. GOA carries two IMP annotations to FTO sourced from PMID:26287746:
GO:0010883 regulation of lipid storage and GO:0090335 regulation of brown fat cell differentiation. Neither is a perturbation of the FTO protein; both are enhancer/IRX3/IRX5
experiments at the FTO locus. I did not use REMOVE (project rule: do not overrule an
experimental annotation whose full text I have not read — here I did read the full text, but the
conservative action is still available and mouse Fto protein data provide independent, weaker
support for an adiposity role). I set GO:0090335 to MARK_AS_OVER_ANNOTATED and kept
GO:0010883 as non-core (it also has IBA and mouse-orthology IEA support), recording the
locus/protein distinction in both reason fields. If a GO curator revisits this, the cleanest fix
is to re-home the PMID:26287746 annotations to IRX3 and IRX5.

3. Dispute layer 1 — chemistry: hydroxylase or demethylase?

Kaur et al. 2025 (NAR) compared FTO with ALKBH5, ALKBH2/3 and bacterial AlkB by MS and real-time
NMR. They open by noting the field is not settled
PMID:40874592 and conclude that the enzymatic product of
FTO is the hemiaminal, N6-hydroxymethyladenosine (hm6A), not adenosine
PMID:40874592. Demethylation happens, but off-enzyme and slowly
PMID:40874592. Their full-text summary is unambiguous
PMID:40874592.
They also flag that standard mapping/quantification workflows may destroy the real product
PMID:40874592.

How much should this move GO? My reading:

• It does not invalidate GO:0016706 2-oxoglutarate-dependent dioxygenase activity. That term
  states the cofactor chemistry only and is the one MF assertion on this gene that no side of
  either dispute contests. I kept it as an accepted, dispute-proof statement of catalytic
  identity.
• It does strain the definitions of GO:0035515 oxidative RNA demethylase activity
  ("Catalysis of the removal of a methyl group...") and GO:1990931 mRNA N6-methyladenosine dioxygenase activity ("...releases oxidized methyl group on N6-methyladenosine as
  formaldehyde"), because under Kaur's model neither the methyl removal nor the formaldehyde
  release is enzyme-catalysed. But the caveat "at least in isolated form" is load-bearing: in
  cells the measured endpoint is loss of the methyl mark, seen repeatedly by independent groups
  PMID:22002720. I therefore did
  not downgrade the demethylase terms on chemistry grounds; I recorded the mechanism dispute in
  reason and proposed a new MF term for the hydroxylase (hm6A-forming) activity so the
  distinction can be curated if it is confirmed in cells.

There is no GO term for m6A/m6Am hydroxylase activity and none for N6-hydroxymethyladenosine
formation (checked via QuickGO search, 2026-09). Proposed in proposed_new_terms.

4. Dispute layer 2 — substrate: mRNA m6A, or snRNA/mRNA-cap m6Am and tRNA m1A?

Solid, uncontested activities.

• snRNA / U6: PMID:30197295
• tRNA m1A, with a striking magnitude contrast against mRNA marks
  PMID:30197295 and a functional consequence
  PMID:30197295
• 5' cap m6Am: PMID:28002401
• The Jaffrey lab's own summary of its prior work
  PMID:41279954
• Ancestral AlkB-type ssDNA/ssRNA alkylation-lesion activity, in vitro
  PMID:18775698, strictly single-stranded
  PMID:18775698. UniProt flags in vivo relevance as "unsure"; I kept the DNA/RNA-repair terms but as
  non-core.

The contested claim: internal mRNA m6A as a physiological substrate.

The classic support is real in vitro activity plus cellular m6A dot-blot/LC-MS changes
PMID:22002720 and nuclear-speckle colocalization
PMID:22002720.
Additional experimental hooks: heat-shock 5'UTR m6A protection
PMID:26458103; inhibitor
chemistry premised on m6A demethylation
PMID:25452335; and structure-guided probes
PMID:26457839.

The counter-evidence is Nicholson et al. 2025, a bioRxiv preprint (not peer reviewed). Using
direct nanopore RNA sequencing — quantitative and single-nucleotide, unlike MeRIP — they report
PMID:41279954, while the same cells show the expected snRNA m6Am
response PMID:41279954 — i.e. an internal positive control that FTO really was
depleted. Their methodological objection to the AML literature is specific
PMID:41279954 and their conclusion is a call to reinvestigate rather than a
retraction PMID:41279954.
They further report the FTO inhibitor FB23-2 is cytotoxic in FTO-null cells, which if it holds
undercuts a chunk of the pharmacological argument.

Against that, the m6A-eraser model remains the working hypothesis of an active drug-discovery
literature published the same year: a structure-based dual-competitive inhibitor
PMID:41190354,
whose prodrug "suppressed AML cell viability, reduced m6A levels, downregulated c-Myc and CEBPA,
and upregulated ASB2 and RARA"; and a degrader
PMID:40815637. Note that the degrader paper
is orthogonal to the inhibitor-specificity objection (degradation, not active-site occupancy),
which makes it the stronger of the two for the pro-m6A side — though it still measures m6A by
antibody-dependent methods.

How much should a preprint move a curation call? — my explicit position.

A preprint should not by itself flip an annotation supported by multiple independent experimental
records, and I did not use REMOVE anywhere on the strength of PMID:41279954. But "not decisive"
is not "not admissible", and three features make this one weigh more than a typical preprint:

1. It is methodologically orthogonal, not a failed replication. Nanopore direct-RNA
   stoichiometry does not share MeRIP's principal failure mode (peak-height variability), which
   is exactly what is being challenged.
2. It carries an internal positive control — the snRNA m6Am increase — that independently
   confirms FTO activity was abolished. A null result with a working positive control is much
   stronger than a bare null.
3. It converges with pre-existing, peer-reviewed observations from the same and other labs that
   FTO's preferred cellular substrate is m6Am rather than m6A (PMID:28002401; PMID:30197295's
   magnitude contrast; PMID:40162895's quantitative m6Am landscape), and with a peer-reviewed
   chemistry paper arriving from an unrelated direction (PMID:40874592).

The appropriate response is therefore demotion, not deletion: GO:1990931 is kept on the gene
(the in vitro activity is not in doubt) but marked non-core, with the dispute written into
reason; GO:1990984 tRNA demethylase activity, the snRNA role, and the generic
GO:0035515 oxidative RNA demethylase activity carry the core function. GOA already hedges the
substrate question by carrying both GO:1990931 (mRNA) and GO:1990984 (tRNA); this review makes
that hedge explicit rather than inventing it.

5. Localization

Nucleus is the dominant and best-supported compartment (EXP/IDA/ISS/IBA/IEA rows), with nuclear
speckles (PMID:22002720) and a genuine cytoplasmic pool that is cell-line dependent
PMID:30197295. All CC
annotations accepted; none of them is in dispute.

6. Term-level notes

• GO:0016180 snRNA processing (IDA, PMID:30197295) — the experiment shows snRNA modification
  (demethylation of m6A/m6Am in snRNAs), not conversion of a primary transcript to a mature
  snRNA. MODIFY to GO:0040031 snRNA modification (a child of GO:0016180) and
  GO:0035513 oxidative RNA demethylation.
• GO:0044065 regulation of respiratory system process (IEA from mouse orthology) — a distal
  organismal phenotype of an Fto-null animal, not something the protein does. MARK_AS_OVER_ANNOTATED.
• GO:0008198 ferrous iron binding (IDA, PMID:20376003) — correct cofactor binding, but it is the
  cofactor rather than the function; non-core. The cached record for PMID:20376003 is abstract
  only and the abstract does not state iron binding explicitly, so no supporting_text is quoted
  for this row; the assignment rests on the curator's reading of the structure paper.
• GO:0035516 broad specificity oxidative DNA demethylase activity — retained non-core. Note FTO
  is not fully "broad specificity" in the sense of the term definition: UniProt records no
  activity towards 1-methylguanine and none towards double-stranded DNA.

7. Open questions carried into suggested_questions

1. Is hm6A (or hm6Am) detectable in cellular RNA, and is it the physiological FTO product?
2. Does any quantitative, antibody-free method detect FTO-dependent internal mRNA m6A changes in
   any cell type?
3. If FTO inhibitors/degraders kill AML cells largely FTO-independently or via snRNA m6Am rather
   than mRNA m6A, what is the actual target engagement?
4. Should GO split m6Am (cap and snRNA) demethylase activity from mRNA m6A dioxygenase activity?

8. Status

Validated with uv run --no-dev ai-gene-review validate --verbose --terms genes/human/FTO/FTO-ai-review.yaml. All PENDING actions replaced; status: COMPLETE.