IFIH1 (MDA5) review notes

Why this gene was selected

MDA5 is annotated in GOA with GO:0038187 pattern recognition receptor activity (IDA x2), a term
whose GO definition is "Combining with a pathogen-associated molecular pattern (PAMP), a structure
conserved among microbial species to initiate an innate immune response." A 2026 Nature Immunology
paper reports that during actual virus infection MDA5 is bound not to viral RNA but to host RNA. The
curation question is whether the PAMP-based molecular function term still fits.

The new result (PMID:42581186, Sampaio et al., Nat Immunol 27:1815-1828, 2026)

Verified on PubMed; DOI 10.1038/s41590-026-02614-3; full text cached.

The authors open by conceding that the ligand has never been pinned down:
PMID:42581186

iCLIP of endogenous MDA5 (new monoclonal antibodies) in THP1 and Calu-3 cells infected with EMCV or
SARS-CoV-2:
PMID:42581186, and more starkly
PMID:42581186 Approximately 10% of input reads in SARS-CoV-2-infected cells mapped to the viral genome, yet
there was no enrichment of viral RNA in the MDA5 IP.

The binding sites were intronic and repeat-proximal, and the trigger was removable by fixing splicing:
PMID:42581186.

Proposed reframing:
PMID:42581186 and, in the discussion,
PMID:42581186.

They describe MDA5's status in the field as unresolved:
PMID:42581186.

Crucially, this is a reframing and not a head-to-head rebuttal. The authors say so themselves:
PMID:42581186. Two cell types, two viruses, one lab.

What the GOA annotations actually rest on

• GO:0038187 IDA #1 = PMID:21217758 (Zuest et al., Nat Immunol 2011, verified). Coronaviruses lacking
  2'-O-methyltransferase induce more interferon, and
  PMID:21217758. Note the logic: this is discrimination by
  absence of a host mark, i.e. missing-self, not detection of a microbe-specific structure.
• GO:0038187 IDA #2 = PMID:23273991 (Wu et al., Cell 2013, verified; cached entry is abstract-only).
  PMID:23273991. Long duplex RNA is a classic PAMP, so this one does support the term as written.
• GO:0003725 dsRNA binding: IBA + IDA (PMID:19656871) + TAS. Uncontested.

Independent evidence that MDA5 senses self RNA

• IFIH1 gain-of-function alleles cause sterile interferonopathy:
  PMID:24686847, with the mechanism
  PMID:24686847.
• ADAR1 A-to-I editing exists in part to stop MDA5 responding to endogenous duplexes:
  PMID:26275108.
• The 2026 paper ties these together:
  PMID:42581186.

Curation position taken

• GO:0003725 dsRNA binding -> ACCEPT, core. Not in dispute under either model; the host ligands
  identified (inverted Alu repeats, structured introns) are themselves duplexes.
• GO:0038187 pattern recognition receptor activity -> ACCEPT, core, with the dispute recorded in
  full in reason. Not REMOVE (two IDAs whose full text I have not read, and the Wu 2013 duplex result
  genuinely is PAMP-like), not MODIFY (no existing MF term describes a receptor for aberrantly
  processed self RNA). The mismatch is with the wording of the GO term, not with this gene, so it is
  raised as a proposed_new_terms entry ("aberrant cellular RNA sensor activity") and a
  suggested_questions item rather than being settled by an edit here.
• GO:0003677 DNA binding (IEA) -> REMOVE. Traced to source: GO_REF:0000002 is InterPro2GO, and
  the InterPro API confirms IPR006935 (Helicase/UvrB, N-terminal; Pfam PF04851 ResIII) maps to
  GO:0003677, GO:0005524 and GO:0016787. That domain family is dominated by DNA-acting enzymes (UvrB,
  type III restriction endonucleases). MDA5 is a cytosolic RNA sensor with no reported direct DNA
  binding; activation by DNA viruses is accepted to be indirect, via RNA. This is exactly the case the
  project rules allow REMOVE for: an electronic inference arguable against on biological grounds.
• GO:0016787 hydrolase activity (same InterPro source) -> MODIFY to GO:0016887.
• GO:0003724 RNA helicase activity -> MODIFY to GO:0008186 (ATP-dependent activity, acting on
  RNA). The GO definition asserts ATP-driven unwinding of an RNA helix. The IMP behind it
  (PMID:19211564) measured ATP hydrolysis and signalling, and reported
  PMID:19211564. The ATPase drives filament proofreading, not processive
  unwinding. GO:0016887, which the gene already carries with EXP/IMP evidence, states what was measured.
• GO:0009597 detection of virus -> KEEP_AS_NON_CORE. The BP term most exposed to the iCLIP result.
• GO:0005634 nucleus (IEA from the UniProt location vocabulary) -> MARK_AS_OVER_ANNOTATED.
  UniProt's own nucleus call for IFIH1 is ECO:0000305 and reads "May be found in the nucleus, during
  apoptosis".
• GO:0016925 protein sumoylation (involved_in) -> MARK_AS_OVER_ANNOTATED; MDA5 is the substrate,
  not a component of the conjugation machinery.
• GO:0005739 mitochondrion -> KEEP_AS_NON_CORE; conditional, post-activation relocalisation to
  engage MAVS, not a resident location.
• Bare GO:0005515 protein binding and GO:0019904 protein domain specific binding ->
  MARK_AS_OVER_ANNOTATED. GO:0042802 identical protein binding kept as ACCEPT, because
  homotypic filament/CARD oligomerisation is mechanistically load-bearing for this protein rather than
  an incidental interaction.

What I could not resolve

Whether the iCLIP result and the in vitro viral-ligand results are in conflict at all. Purified MDA5
demonstrably binds long viral duplexes; iCLIP reports what MDA5 is crosslinked to inside an infected
cell, where viral RNA may be shielded in replication organelles. The authors themselves argue this is
the resolution rather than a contradiction. I did not find a 2025-26 paper that tests the two models
against each other directly.

2026-10-25 ORGANELLES batch 3 (mitochondrion rows)

Rows: GO:0005739 mitochondrion, IDA PMID:33727702 and IEA GO_REF:0000044 (SL-0173). No sub-compartment rows; qualifier located_in.

Checks:
- UniProt "Mitochondrion {ECO:0000269|PubMed:33727702}. Note=Upon viral RNA stimulation and ISGylation, translocates from cytosol to mitochondrion." [file:human/IFIH1/IFIH1-uniprot.txt].
- PMID:33727702 (full text cached): kit-based fractionation, 10,000 g pellet; "Anti-α-tubulin and anti-MAVS immunoblotting served as markers for the cytosolic and mitochondrial fractions, respectively." No ER marker, and MAVS is also on MAM and peroxisomes, so the pellet does not separate mitochondrion from MAM.

Outcome: both rows stay KEEP_AS_NON_CORE; the reason no longer says "outer membrane", which the paper does not show; YAML anchor on supported_by replaced by explicit lists. Stays at root (TRAF3 precedent).