AIRIM / C1orf109 (Q9NX04) — review notes

2026-09-26. PAINT no-IBA backlog. Deep research provider: affinage.

Shape of the problem

188 GOA rows, but only 10 distinct GO terms. 174 of the rows are GO:0005515 protein binding
IPI, one per interaction partner (148 distinct partners after the seeder's known WITH/FROM
collapse merged a duplicate CINP row). Where they come from:

So ~164 rows come from three systematic screens that were not studies of AIRIM at all.

How the protein-binding mass was handled

Project policy (CLAUDE.md + the annotation-reviewer skill) is explicit and I followed it rather
than inventing a scheme: GO:0005515 is not MARK_AS_OVER_ANNOTATED — its defect is that it
conveys no functional information, not that it overclaims. MODIFY where a more informative
evidence-backed term exists, otherwise REMOVE, and removal does not assert the interaction is
false.

That split fell out cleanly on the partner, not the reference:

• 8 rows MODIFY → GO:1904949 ATPase complex. Partner is CINP, AFG2A or AFG2B — genuine 55LCC
  subunits. Complex membership is the informative statement these rows were groping toward.
  Notably this includes the two BioPlex rows, whose single partner each is CINP, so an AP-MS
  screen row was treated differently from a Y2H screen row purely on who the partner is.
• 166 rows REMOVE, each naming its own partner (resolved via the UniProt API, including 20
  isoform accessions mapped to their base entries — not written from memory).

The real biology

AIRIM is a structural subunit of the 55LCC complex (AFG2A/SPATA5 + AFG2B/SPATA5L1 + CINP +
AIRIM):

PMID:38554706
PMID:38554706

One structural role, two settings:

• Cytoplasmic pre-60S maturation — PMID:35354024, with PMID:35354024
• Chromatin replisome proteostasis — PMID:38554706

The MF gap, and how I filled it without over-reaching

Stripping 166 protein-binding rows would have left the gene with zero molecular function
annotations — which misrepresents a protein whose role is structurally resolved. Added one
NEW: GO:0005198 structural molecule activity ("the action of a molecule that contributes to
the structural integrity of a complex"), which is exactly what the funnel architecture shows.

Critically, AIRIM is not the ATPase — that is the AFG2A/AFG2B motor. The schema anticipates
this case, so the ATPase activity goes in contributes_to_molecular_function (GO:0016887), not
molecular_function. A term for substrate gating at the funnel would be better if one existed;
raised in suggested_questions instead of forced.

Two gotchas worth recording

1. GO:1904949 ATPase complex is rejected in locations. I had put it in both locations
   and in_complex; GOCellularLocationEnum excludes it. Belongs only in in_complex.
2. The --terms validator failure is silent through the wrapper. just validate printed
   ✗ Invalid with no error line at all. Running linkml-term-validator validate-data
   directly gave the actual message immediately. Worth remembering: an "Invalid with no errors
   shown" result means drop to the underlying tool.

The IBA rows are fine

Both IBA rows carry UniProtKB:Q9NX04 — AIRIM itself — in their own WITH/FROM. Per the skill's
guidance this is expected and correct, marking that the node has experimental grounding on
this very gene. Not labelled circular.

One demotion

GO:0005730 nucleolus (IDA) → KEEP_AS_NON_CORE. Plausible for a ribosome factor and a real
observation, but the demonstrated AIRIM step is late and cytoplasmic, and the replisome role is
nucleoplasmic. Neither characterised activity happens in the nucleolus.

affinage assessment

Accurate and well framed this time — correctly names the 55LCC complex, calls AIRIM's role
structural, separates the two activities, and surfaced three papers absent from GOA
(PMID:40268917, PMID:40760247, PMID:32761833). A useful contrast with ADIG (missed the decisive
paper) and ADM5 (right papers, wrong framing) earlier in the same session.

Outcome

189 rows: 166 REMOVE, 13 ACCEPT, 8 MODIFY, 1 KEEP_AS_NON_CORE, 1 NEW.
just validate human AIRIM → ✓ Valid, no warnings.
17 distinct supporting_text quotes (358 instances) verified as substrings, including the 2
file: ones CI does not check.

Round 2 (review feedback)

Two blocking items, both fair, and the first is a pattern I have now hit three times in one
session.

Three AIRIM-specific papers were in my own diff and uncited

PMID:40268917, PMID:40760247 and PMID:32761833 — 37, 73 and 105 AIRIM mentions
respectively — were cached by this PR and cited nowhere. Worse, the description asserted that
hypomorphic AIRIM variants cause a neurodevelopmental disorder on the strength of nothing,
while the source sat in the same diff:

PMID:40760247

Same shape as the ADIG cold-tolerance claim and the ADM5 heart-failure diuresis claim earlier
today. The common cause is that I cache papers while researching, use them to form the picture,
then write the picture without going back to cite what formed it. The three-time recurrence means
it is a process failure, not a slip.

"Substrate presentation" was wrong

I wrote that AIRIM's contribution is "structural integrity and substrate presentation". The
higher-resolution structure says the opposite:

PMID:40268917
PMID:40268917

AIRIM makes no direct pre-60S contact. The structural-integrity half stands and GO:0005198
is unaffected — the N-terminal ring supports it — but the substrate half is removed, and the
suggested_questions entry that speculated about AIRIM gating access to the motor is recast to
say explicitly that gating is ruled out.

Stoichiometry

PMID:40268917
PMID:40268917

"Heterohexameric" (which I took from UniProt and PMID:38554706) describes the SPATA5–SPATA5L1
motor, not the 4:2:2:2 assembly. Replaced everywhere it described the whole complex; the three
surviving occurrences are the sentences explaining the distinction. "Funnel" is likewise replaced
by "N-terminal ring", which is what the structure paper calls it.

Isoform caveat recorded

PMID:32761833 is about the long isoform C1orf109L binding DHX9 and promoting R-loop
dependent DNA damage. Recorded as a flagged caveat rather than folded in: it is a different
activity from the 55LCC roles, no GO annotation on this gene covers it, and it should not be
assumed to hold for the canonical product.

Not changed

The six MODIFY rows still propose GO:1904949, which the gene already carries. That is
deliberate — the point of the MODIFY is that these protein-binding rows should become the
complex-membership statement, and duplicates to the same term are explicitly fine per the
annotation-reviewer guidance. Also unchanged: the NEW row's file: evidence leg is skipped by
the supporting-text validator, which is true of every file: quote in this repo; I verify them
by hand and say so.

Final: 189 rows, 166 REMOVE / 13 ACCEPT / 8 MODIFY / 1 KEEP_AS_NON_CORE / 1 NEW.
25 distinct quotes (366 instances) verified.

Round 3 — I over-corrected on round-2 feedback, and the reviewer caught their own error

Round 2's reviewer told me PMID:40268917 refuted "substrate presentation", citing "C1orf109 does
not interact with the pre-60S particle directly". I removed the claim. In round 3 the same
reviewer retracted that: the sentence is about the docking platform, not the substrate.

I verified the retraction rather than accepting it on the rebound, and it is right. The AAA+
substrate is RLP24, and on that the paper concludes the opposite of what I had written:

PMID:40268917

It is not a structural inference alone — there is co-IP:

PMID:40268917

and the structure names the binding element:

PMID:40268917

with AIRIM forming the surface it binds against:

PMID:40268917

The accurate distinction is narrow: AIRIM binds the substrate RLP24 inside the funnel, but
makes no direct contact with the pre-60S particle the substrate sits on — docking onto the
particle is CINP's job. My round-2 text collapsed those two into "AIRIM does not present
substrate", which is false.

Consequence: a better molecular function

The paper calls C1orf109 an adapter protein outright, so GO:0060090 molecular adaptor activity is directly supported — "bringing together two or more molecules through a selective,
non-covalent interaction, permitting those molecules to function in a coordinated way" is exactly
the RLP24/motor relationship. Added as a second NEW row and promoted to
core_functions.molecular_function.

GO:0005198 stays alongside rather than being replaced: holding the NTD ring together and
delivering the substrate are separable claims, and either could be true without the other.

Also

• RLP24 was named nowhere in the review despite being the substrate the whole complex acts
  on. Now named 13 times.
• Review-process commentary removed from core_functions ("an earlier draft…", "is avoided
  here"). Same fix as ADM5 round 4 — a curation field should read as curation.
• "Funnel" restored. I had replaced it with "N-terminal ring" thinking it was loose language;
  both papers use "funnel", and the structure paper's own phrase is "the inner wall of the
  funnel".

The lesson

Round 2's correction was made on a reviewer's citation without my re-reading the surrounding
passage. The citation was real and the inference from it was wrong, and I propagated that into
two fields. A quote handed to me needs the same context check as a quote I find myself —
arguably more, because an authoritative-sounding correction invites less scrutiny than my own
draft does.

Final: 190 rows, 166 REMOVE / 13 ACCEPT / 8 MODIFY / 1 KEEP_AS_NON_CORE / 2 NEW.
29 distinct quotes (370 instances) verified.

Round 4 (review feedback)

The retraction had not reached the reference block

I corrected core_functions and suggested_questions in round 3 but left the error standing in
references[PMID:40268917](https://pubmed.ncbi.nlm.nih.gov/40268917), where it did the most damage — a findings[].statement asserting as
a finding of the paper the opposite of that paper's own section heading:

"C1orf109 does not interact with the pre-60S particle directly, which rules out a
substrate-presentation role for AIRIM."

and review_notes carrying the round-2 conflation in compressed form ("CINP makes both
interfaces" — those interfaces are with the particle, not the substrate) plus "the substrate
half does not survive", contradicted by this file's own GO:0060090 row a hundred lines above.

Both corrected, and the adapter/RLP24 conclusion added to that reference's findings, where it had
been missing despite now grounding a NEW MF row.

This is the fourth time in this session a retraction has failed to reach every field that
repeats the claim (ADM5 summary vs reason; ADM5 description and suggested_questions; here,
twice). The pattern is consistent enough to name: a correction is not done when the sentence
that prompted it is fixed — it is done when every field asserting the claim has been grepped.

The reviewer offered finding_review with finding_status: OVERTURNED + superseded_by as a
way to preserve the error. I did not use it, because that mechanism is for a paper's finding
being refuted by later work. Here the paper was right throughout and my statement about it was
wrong. Marking the paper's finding overturned would misattribute my error to it.

Smaller items

• GO:0005198's reason said a substrate-gating term "would be better if one existed" — one now
  does, thirty lines below. Rewritten to point at GO:0060090 and explain why both are proposed.
• "Kept alongside" was true of the proposals but not of core_functions, where molecular_function
  is single-valued and GO:0060090 took the slot. Now says so explicitly.
• Adaptor row evidence code changed IDA → IPI: the load-bearing experiment is a physical
  interaction (the RLP24 fragment pulling down C1orf109), not a direct assay of the activity.
• Two quotes used ASCII hyphens where the cached record has en-dashes (residues 85–163,
  residues 80–163). Local validation passed because normalize_text strips punctuation, so this
  was invisible to the gate — but a "verbatim quote" should be verbatim. Fixed to the real
  characters.

History record

Added, covering all four rounds. Missing for the same reason as AJM1: CI does not enforce it.