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.