AAMDC (Mth938 domain-containing protein) — review notes

PAINT no-IBA project review, using the affinage deep-research provider
(AAMDC-deep-research-affinage.md, gates passed) plus UniProt Q9H7C9, the GOA TSV and the
primary literature.

Headline: the IBAs here are good, and the protein binding set is worthless

This gene is the near-mirror-image of A1BG (reviewed earlier in this campaign), which makes
it a useful calibration point. There, the IBAs were wrong and needed removal. Here they are
the only annotations with real evidence behind them, and the experimental-looking IPI
annotations are the ones that should be discounted.

The two IBAs both trace to a genuine mouse ortholog with primary data

Both IBAs cite MGI:MGI:1913523, which is mouse Aamdc — the true ortholog, not a
paralog. Its own GO record carries, from PMID:22279136:

• GO:0045600 positive regulation of fat cell differentiation — IDA and IMP
• GO:0045944 positive regulation of transcription by RNA polymerase II — IDA
• GO:0043066 negative regulation of apoptotic process — IMP

and GO:0005737 cytoplasm by IDA from PMID:21622130.

Reading the source paper confirms the adipogenesis annotation is well-founded and
bidirectional — gain of function and loss of function
[PMID:22279136, "Our results indicated that LI2 was sufficient to drive preadipocyte
differentiation via modulating the phosphorylation level and transcriptional activity of
CREB"] and [PMID:22279136, "knockdown of the LI2 protein resulted in preadipocyte apoptosis
via caspase-3 activation during adipogenesis"]. (The mouse gene was called LOC66273
isoform 2 / "LI2" in that paper; hence the "adipogenesis associated" in the human gene name.)

So GO:0045600 is exactly what IBA is for: a solid ortholog phenotype transferred to a human
gene with no experimental data of its own. ACCEPT. UniProt's own FUNCTION line for human
AAMDC is ECO:0000250 (by similarity) for the same reason
[file:human/AAMDC/AAMDC-uniprot.txt, "May play a role in preadipocyte differentiation and"].

The eight protein binding IPIs are all high-throughput, and five are a classic artefact

The five-partner block from PMID:32814053 is the giveaway. That paper is "Interactome
Mapping Provides a Network of Neurodegenerative Disease Proteins and Uncovers Widespread
Protein Aggregation in Affected Brains" — but note that title phrase refers to
aggregation seen in postmortem patient brain tissue, not to the interaction dataset. The
argument against these five rests on the screen design instead: one systematic Y2H pass over
~500 neurodegeneration-related baits, with a single small uncharacterised protein scoring
against five of them and no independent replication or follow-up for any pair. AAMDC has no
described role in neurodegeneration.

The remaining three are single-publication Y2H hits (UniProt's NbExp=3 counts assay
replicates within one study, not independent studies).

The most telling observation: not one of the eight recovered RABGAP1L or RAB7A — the only
AAMDC interaction with functional follow-up in the literature. The GOA binding record for this
gene is entirely orthogonal to its known biology. All eight are marked over-annotated.

What the human literature actually shows

One substantial human paper, PMID:33772001 (Nat Commun 2021), studying AAMDC as an oncogene in
the 11q13.5–14.1 (IntClust2) amplicon of ER+ breast cancer:

• AKT activation is causal, not correlative —
  [PMID:33772001, "Ectopic AAMDC expression is sufficient to activate AKT signaling, resulting
  in estrogen-independent tumor growth."] This supports a NEW GO:0051897 annotation, and is
  the only human functional annotation this gene can currently carry.
• PI3K-AKT-mTOR control and metabolic reprogramming —
  [PMID:33772001, "We show that AAMDC controls PI3K-AKT-mTOR signaling, regulating the
  translation of ATF4 and MYC and modulating the transcriptional activity of AAMDC-dependent
  promoters."]
• RABGAP1L/RAB7A endolysosomal platform —
  [PMID:33772001, "we provide evidence that AAMDC can interact with the RabGTPase-activating
  protein RabGAP1L, and that AAMDC, RabGAP1L, and Rab7a colocalize in endolysosomes."] Note
  the authors' own hedge ("can interact", "provide evidence"). Colocalisation is not
  co-residence in a complex, so GO:0036019 endolysosome is proposed but flagged as resting
  on colocalisation from a single study.

The molecular function is genuinely unknown

Worth stating plainly, because it is the main knowledge gap: AAMDC has no molecular function
annotation of any kind, and none is currently justifiable. Mouse Aamdc even carries an
explicit GO:0003674 molecular_function ND (no data). The protein is a small Mth938-domain
protein with a solved structure but no assigned activity, and affinage's mechanism_profile
reports molecular_activity: (none) — the provider agreeing there is nothing to ground. Every
described effect (adipogenesis, AKT activation, ATF4/MYC translation) is a downstream cellular
consequence, not a biochemical activity. I have deliberately not invented an MF term.

Tested, not just asserted

A negative claim deserves evidence, so I ran a family-wide analysis
(AAMDC-bioinformatics/analyze_mth938.py → RESULTS.md; every figure fetched at run time
from the InterPro, UniProt and QuickGO REST APIs, re-runnable). Enumerating every reviewed
Swiss-Prot member of AAMDC's Pfam family PF04430 — aptly named DUF498, "domain of unknown
function":

The missing molecular function is a property of the entire family, not an oversight on
this gene.

And the family points somewhere specific

The family is not wholly uncharacterised. Its other branch is NDUFAF3, with a consistent
function in every organism studied — human, mouse, rat, bovine, zebrafish, Xenopus,
Drosophila — an assembly factor for mitochondrial complex I, the Drosophila entry
experimentally supported (PubMed:34386730).

That is informative in a particular way. NDUFAF3 is an assembly factor, not an enzyme: it
binds subunits, helps build a complex, has no catalytic activity, and is not part of the
finished product. Consistent with that, no member of this family — NDUFAF3 included — has a
catalytic activity block or an experimental MF term.

So the family signal does not suggest a hidden enzymatic activity awaiting discovery. It
suggests a protein-assembly or chaperone-like role — precisely the class of function GO
cannot currently express as a molecular function (see the AAGAB review in this campaign,
where the same gap was filed under proposed_new_terms). Two genes in, that is twice the
same structural limitation in GO.

The right experimental question is therefore probably not "what does AAMDC catalyse?" but
"what does it help assemble?"

Where the localisation bar sits, and why

PMID:33772001 reports three localisation observations; only one is annotated, so the
asymmetry is worth stating.

• Endolysosome (annotated, non-core). AAMDC, RABGAP1L and RAB7A colocalise there. Taken,
  because it is tied to a specific named interaction the paper pursues mechanistically —
  with the caveat now in the annotation that GO:0036019 denotes the transient hybrid
  organelle from late-endosome/lysosome fusion, which IF colocalisation with RAB7A (a broad
  late-endosome/lysosome marker) cannot resolve.
• Nuclear and plasma-membrane staining (not annotated). The same paper reports AAMDC "in
  both cytoplasmic and nuclear compartments, accompanied by plasma membrane staining
  depending on the cell line analyzed". Cell-line-dependent IHC with no functional experiment
  attached, and not pursued by the paper.
• Endosomal trafficking (not annotated). A knockdown lysosomal phenotype is said to
  substantiate "a role of AAMDC in the regulation of endosomal trafficking". Left out: a
  single-sentence interpretation of a phenotype, and a trafficking-regulation annotation on a
  protein with no known molecular function would be exactly the over-reach this review
  otherwise avoids.

The bar: a compartment is annotated when a named, mechanistically pursued interaction puts it
there — not on staining pattern, and not on a phenotype's interpretation alone.

Actions