Reviewed as one half of a paired batch with YTHDF3, because the two genes share a single
contested premise: whether N6-methyladenine (6mA) exists at functionally meaningful levels in
mammalian genomic DNA. ALKBH1 is the proposed eraser; YTHDF3 has recently been proposed as
the reader. If the substrate is largely artifactual, both assignments lose their ground at once.
ALKBH1 (AlkB homolog 1, ABH1) is one of nine mammalian AlkB-family Fe(II)/2-oxoglutarate
dependent dioxygenases. It carries the canonical 2OG-Fe(II) oxygenase fold and the
His-X-Asp...His iron triad. Unlike ALKBH2/ALKBH3, which were straightforwardly shown to be
1-methyladenine/3-methylcytosine DNA repair demethylases, ALKBH1 was initially refractory:
The protein carries an N-terminal mitochondrial targeting signal and is predominantly
mitochondrial, with a smaller nuclear pool:
This is the activity with the strongest, most independently replicated support, and the one I
treat as ALKBH1's core molecular function.
Mammalian mitochondria use a non-universal genetic code in which a single mt-tRNA-Met must
decode AUG, AUA and AUU. That is achieved by a two-enzyme modification cascade at the wobble
position C34: NSUN3 makes m5C34, then ALKBH1 oxidises it to f5C34.
Independently confirmed and mechanistically dissected by the Suzuki lab, who also reconstituted
the two-step oxidation in vitro:
The same paper found ALKBH1 also acts on cytoplasmic tRNA-Leu (hm5Cm/f5Cm at position 34) and
demethylates m1A in mt-tRNAs:
Reinforced in 2025. A paper whose whole point is to deflate an epitranscriptomic claim
nevertheless leaves ALKBH1's mt-tRNA role standing — and in fact strengthens it by removing the
competing mRNA-f5C and TET-enzyme claims:
This is a useful contrast with the 6mA story below: when a rigorous methodological audit is
applied to f5C, the ALKBH1 mt-tRNA assignment survives. When the equivalent audit is applied
to genomic 6mA, the substrate itself largely disappears.
Corroborated independently by the m1A increase seen in ALKBH1-KO mt-tRNAs (PMID:28472312,
quoted above). Two labs, two systems, consistent direction — I treat this as core.
ALKBH1 does demethylate 3-methylcytosine, but only in single-stranded DNA and in RNA:
It also has a second, chemically distinct activity — AP lyase — arising from a different active
site, which the authors themselves hedge about physiologically:
There is also a reported histone H2A dioxygenase activity (not currently in GOA):
And an m3C-in-mRNA demethylase report:
Caveat worth recording: the m3C-in-mRNA premise rests on METTL8 being an mRNA m3C writer, and
METTL8 has since been re-characterised as a mitochondrial tRNA m3C32 methyltransferase. The
molecular function (oxidative RNA demethylase) is not in doubt; the mRNA substrate assignment
is weaker than the tRNA one.
Two 2018 papers underlie GO:0141131 DNA N6-methyladenine demethylase activity (EXP):
The mouse ESC precedent is Wu et al. 2016:
Three independent, technically focused studies concluded that mammalian genomic 6mA measurements
are dominated by artifact:
Musheev et al. is the most damaging for the eraser hypothesis specifically: if genomic m6dA is
mis-incorporated from the ribonucleotide salvage pool rather than written by a methyltransferase,
there is no regulated mark for an eraser to erase, and any apparent "demethylase" effect on bulk
genomic 6mA could instead reflect the enzyme's RNA activity changing the m6A-ribonucleotide pool.
This adds a phylogenetic argument to the analytical one: animals have lost the AMT1 writer
lineage. A dedicated eraser for a mark with no writer is hard to motivate.
The EMBO J 2025 paper that proposes YTHDF3 as the 6mA reader opens by conceding the dsDNA
problem in its own abstract:
Its proposed resolution is that YTHDF3 recruits ALKBH1:
Note that the senior/corresponding authors (Yan G-R, Gu X-F) overlap with those of PMID:30017583,
so this is not an independent confirmation of the substrate's existence — it is a continuation of
the same programme.
GO:0141131 carries EXP codes; CLAUDE.md forbids overruling an experimentalKEEP_AS_NON_CORE for GO:0141131, and likewise for the downstreamGO:0141137 positive regulation of gene expression, epigenetic andGO:0035516 broad specificity oxidative DNA demethylase activity annotations that inheritGOA gives human ALKBH1 GO:0042056 chemoattractant activity (IEA, GO_REF:0000107, Ensembl
Compara transfer from mouse UniProtKB:P0CB42) and, logically inferred from it,
GO:0050918 positive chemotaxis (IEA, GO_REF:0000108).
Tracing the mouse source via QuickGO: UniProtKB:P0CB42 enables GO:0042056 IDA PMID:16860792
(assigned by MGI, 2006). That paper is about NRP (neural regeneration protein), and states:
NRP is a different reading frame product of the Alkbh1 locus — a secreted, SDF-1-like
neurotrophic/chemoattractive factor with predicted trefoil (TFF-1) and survival-promoting-peptide
homology. It is not the ALKBH1 dioxygenase polypeptide. Transferring a secreted-chemokine
molecular function onto the mitochondrial 2OG-Fe(II) oxygenase Q13686 by Compara orthology is an
over-propagation: a mitochondrial-matrix tRNA-modifying dioxygenase with no signal peptide cannot
be a secreted chemoattractant. Both IEA rows are marked REMOVE.
This is exactly the category CLAUDE.md permits for removal — "EC/IEA mappings that are
demonstrably wrong... over-propagated electronic (IEA/IBA) inferences you can argue against on
biological grounds" — and it touches no experimental annotation.
GO:0000791 euchromatin (IEA, Compara from mouse) traces to a genuine Alkbh1-protein resultGO:0005739 mitochondrion is supported by many independent lines (IDA in PMID:18603530,GO:0006446 regulation of translational initiation (IMP) is correct but is the general parentGO:1990983 regulation of translational initiation by tRNA modification,MODIFY.ALKBH1's best-supported core function is not DNA demethylation. It is a mitochondrial
(and partly nuclear/cytoplasmic) 2OG/Fe(II) tRNA-modifying dioxygenase:
Its DNA-directed activities (ssDNA m3C demethylation, AP lyase, and the contested genomic 6mA
demethylation) are retained as non-core.