ZBP1 (Q9H171) review notes

Human Z-DNA-binding protein 1 / DAI / DLM-1. Reviewed as a contested-function case:
two separable disputes sit on top of a GOA record that is unusually thin in direct human
experimental evidence.

Dispute 1 — can the Zα domains convert dsRNA to Z-form?

The load-bearing GO row is GO:7770073 left-handed Z-RNA immune receptor activity, and
in human GOA it is carried only as ISS from mouse Zbp1 (UniProtKB:Q9QY24,
GO_REF:0000024). GO:0003692 left-handed Z-DNA binding is carried only as IBA and
NAS (PMID:11842111). For the two functions the gene is named for, neither has a
direct human experimental code in GOA. That is a real curation weakness independent of the
scientific dispute.

The 2026 challenge (a PREPRINT — weigh accordingly). Krall et al., bioRxiv 2026
(PMID:42079158, publication_type: PREPRINT) point out that the assumed A-to-Z RNA
conversion activity was inherited from ADAR1 by analogy and never tested:
PMID:42079158

By CD, NMR and ITC they find:
PMID:42079158 but
PMID:42079158,
attributing the loss to the missing cis-trans proline-proline motif in both ZBP1 Zα domains.

Crucially, this is a refutation of a mechanism, not of the receptor function. The same
preprint shows ZBP1 binds and stabilises Z-RNA once the energetic barrier is lowered:
PMID:42079158 and
PMID:42079158. Their own conclusion is a
narrowing of substrate scope, not an abolition of sensing:
PMID:42079158.

The GO definition of GO:7770073 is "Combining with a left-handed Z-RNA and transmitting
the signal to initiate an innate immune response" (checked against the GO API). It requires
binding + signal transduction, not A-to-Z conversion. So the preprint does not refute the
term as written; it refutes a mechanistic story often told alongside it.

The other side. Yin et al., Nature 2025 (PMID:41082924) — the paper the Sci Bull
commentary PMID:41580360 is a Comment on — provide the positive cellular case that
physiological Z-RNA ligands exist:
PMID:41082924 and
PMID:41082924. These arise from disruption of transcription termination, i.e. they are
generated by the cell under torsional/processing stress — exactly the "pre-Z-forming"
category the preprint says ZBP1 can read. The two papers are therefore more compatible
than the framing suggests.

Earlier, directly relevant evidence: influenza-generated Z-RNA activates ZBP1 in the nucleus
PMID:32200799, and vaccinia E3 blocks ZBP1 by competing for Z-RNA
PMID:34192517.

Z-DNA binding by human ZBP1 is, by contrast, directly demonstrated biophysically and has
been since 2006 — the NAS/IBA codes in GOA understate the evidence:
PMID:16990255.

Position taken: GO:7770073 ACCEPT (core), GO:0003692 ACCEPT (core), with the
conversion dispute and the ISS/NAS/IBA evidence weakness recorded in each reason, and an
explicit recommendation that both be upgraded to direct human experimental evidence.
UNDECIDED was considered and rejected: the term's definition is about receptor activity,
the cellular evidence for Z-RNA-triggered ZBP1 activation is strong and from several labs,
and a single preprint refuting an adjacent mechanistic claim does not make the receptor
assignment unresolvable.

Dispute 2 — does human ZBP1 signal like mouse ZBP1?

Lu et al., EMBO Rep 2026 (PMID:42436309):
PMID:42436309
PMID:42436309
PMID:42436309

How far this warning actually reaches. Being careful here matters. The paper reports that
hZBP1 is a more potent inducer of cell death than mZBP1, not a weaker one. What diverges is
the downstream kinase requirement and RHIM usage, not the existence of the cell-death
output. So the mouse-derived BP terms describing the output — positive regulation of
necroptotic process, necroptotic signaling pathway, positive regulation of apoptotic process,
defense response to virus — survive intact, and in fact the paper strengthens them for human.
Per CLAUDE.md, the IBA question is whether human ZBP1 sits outside the clade that inherited
the function or has diverged in the relevant respect. It does not: it is the 1:1 ortholog,
it performs the same process, and the PAINT node is seeded partly by human ZBP1's own
experimental annotations. The IBAs are accepted.

The paper also weakens, rather than strengthens, a purely Zα-driven model:
PMID:42436309 — although Zα mutations do attenuate:
PMID:42436309.

Where the warning does bite is the type I interferon arm, and there it converges with
much older human-specific data (see below).

The type I interferon arm — GO:0060340

Two rows: IBA (GO_REF:0000033) and IEA from InterPro:IPR042361 (the ZBP1 family entry,
whose InterPro description still reads "a cytosolic DNA sensor that activates the innate
immune system"). The mouse PAINT seed for this term is an IMP from PMID:17618271
(Takaoka 2007, the original DAI paper) — confirmed by querying mouse Zbp1's annotations.

Two independent problems:

  1. Term scoping. GO:0060340 is "positive regulation of type I interferon-mediated
    signaling pathway" — regulation of signalling downstream of the IFN receptor. What
    Takaoka reported is DAI-dependent induction of IFN genes, i.e. IFN production
    (GO:0032481). The term does not describe the claimed biology.
    PMID:17618271
  2. Species-specific refutation. Lippmann et al. 2008 tested this directly in human cells:
    PMID:18771559 and
    PMID:18771559 — while mouse Zbp1 siRNA did impair
    the response in L929 cells. This is exactly the shape of divergence the EMBO 2026 paper
    warns about, reported eighteen years earlier.

Position taken: MARK_AS_OVER_ANNOTATED on both GO:0060340 rows. Not MODIFY to
GO:0032481, because asserting that human ZBP1 positively regulates type I IFN production
would be asserting the thing Lippmann specifically failed to find in human cells. The IBA
carries a propagation_review with root_cause: TERM_SCOPING_PROBLEM; I am not challenging
the PAINT node placement, only the term chosen and its transfer to human.

GO:0003726 — dsRNA adenosine deaminase activity (IEA)

Present, and wrong. Traced it: the annotation comes from InterPro:IPR042371, which is the
generic "Z-binding domain" entry, and that entry carries GO:0003723 and
GO:0003726 (verified via the InterPro API). The deaminase mapping is inherited from ADAR1,
the prototypical Zα-domain protein and an actual adenosine deaminase. ZBP1 has no deaminase
domain at all — its architecture is Zα1–Zα2 followed by three RHIMs
PMID:42079158, and the
UniProt record for Q9H171 lists no deaminase domain or catalytic residues. This is a
textbook domain-driven electronic mis-mapping: a shared accessory domain dragging along the
catalytic activity of a different protein. REMOVE.

(The sibling mapping on the same InterPro entry, GO:0003723 RNA binding, is fine.)

Best-supported core function

Honest reading of the whole record: the most robustly evidenced thing human ZBP1 does is
RHIM-dependent assembly of a death-signalling complex. That has direct human experimental
support going back to 2009 —
PMID:19590578 — and is the entire subject of the 2026 human-vs-mouse paper. GOA represents
it only as bare GO:0005515 protein binding, which carries no functional information. A
NEW annotation of GO:0035591 signaling adaptor activity is proposed, and the review
places three core functions: the Z-RNA receptor arm, the RHIM adaptor arm, and Z-DNA
binding, with the adaptor arm the best-evidenced of the three in human cells.

Note on the Z-RNA core function's term id: GO:7770073 is used on the annotation row, but
core_functions ids are hard-validated against the repo's ontology snapshot, and the term was
only created on 2026-05-07 (GO issue #32046), so it is not yet in that snapshot. The core
function therefore declares the parent GO:0038187 pattern recognition receptor activity and
names the specific child in its description. This produces one validator warning (the parent is
not itself in existing_annotations), which is preferred here over either failing validation or
adding a redundant parent-term annotation to the GOA-derived block.

Incidental flags

Summary of what changed

Term Action Why
GO:7770073 Z-RNA immune receptor ACCEPT (core) definition is binding+signalling, not conversion; ISS-only evidence flagged
GO:0003692 Z-DNA binding ACCEPT (core) directly demonstrated for human protein; IBA/NAS understates it
GO:0003726 dsRNA adenosine deaminase REMOVE Zα-domain-driven InterPro mis-mapping from ADAR1
GO:0060340 pos reg type I IFN signalling MARK_AS_OVER_ANNOTATED ×2 wrong term scope + human-specific negative result
GO:0003677 DNA binding MODIFY → GO:0003690 ×2 the experiment measured dsDNA binding
GO:0002376 immune system process MODIFY → GO:0140374 uninformative root-level term
GO:0050727 reg inflammatory response MODIFY → GO:0050729 direction is known
GO:2000659 reg IL-1 signalling MARK_AS_OVER_ANNOTATED wrong arm of IL-1 biology; mouse-only
GO:0005515 protein binding ×3 MARK_AS_OVER_ANNOTATED project guidance
GO:0035591 signaling adaptor activity NEW the best-evidenced human function, unrepresented