CASP4 (human caspase-4, P49662) — review journal

Reviewer: AI curation pass, 2026-09-17. 99 GOA annotations reviewed.

1. What CASP4 is

Human CASP4 (originally cloned as ICH-2 / TX / ICE(rel)-II in 1995) is an inflammatory
caspase of the caspase-1 subfamily, encoded in the CASP1/CASP4/CASP5 cluster on 11q22.
It has an N-terminal CARD followed by a p20/p10 catalytic protease domain with the
canonical cysteine nucleophile, and it cleaves after Asp at P1.

The founding papers already showed it is a bona fide protease that is not an IL-1β
converting enzyme in the classical sense:
PMID:7797510 and
PMID:7743998

Its defining modern role is as the effector of the non-canonical inflammasome: it is a
cytosolic receptor for bacterial LPS that, on engaging LPS, oligomerizes, autoprocesses,
and cleaves GSDMD to drive pyroptosis
PMID:26375003.
Autoprocessing is mechanistically required, not incidental
PMID:32109412 and
PMID:37558421.

It also directly matures pro-IL-18, which two independent 2023 structures established
PMID:37993714
PMID:37993712.

2. Contested question A — what does CASP4 actually engage?

GOA carries GO:0001530 lipopolysaccharide binding (IDA, PMID:25119034 and PMID:37993712)
and its parent GO:0008289 lipid binding (IDA ×3). The IDA rests on the 2014 Nature paper
PMID:25119034, i.e. a direct, high-affinity molecular interaction with lipid A.

The 2026 Broz-lab paper reframes the ligand as a membrane surface with geometry:
PMID:41702406 and
PMID:41702406, with GBP1 an obligate upstream mechanoenzyme
PMID:41702406.

My reading: these are not contradictory at the level GO annotates. The same paper still
calls lipid A the cognate ligand PMID:41702406, and the concurrent Sci Adv study
localises the chemical recognition event to a lipid-binding pocket in the CARD
PMID:41477831.
The curvature result constrains presentation and avidity (how many lipid A moieties are
accessible, and where), not whether the CARD chemically binds LPS. It is also entirely
consistent with the older GBP1 work, which had already framed GBP1's job as unmasking lipid A
PMID:32510692 and
PMID:31268602; and with 2026 work putting caspase-4 recruitment at LPS-bearing
surfaces of extracellular bacteria PMID:42265284.

Decision: ACCEPT GO:0001530 (do not MODIFY, do not REMOVE), and record the curvature
caveat in the annotation reason and in suggested_questions. GO has no term for "binding
to a curved lipid membrane surface", and inventing one on a single paper would be premature
— a membrane curvature sensing style term would be the thing to propose if the result is
replicated. GO:0008289 lipid binding is MODIFY→GO:0001530: all three IDA rows derive from
LPS/lipid A binding experiments and the specific child is already present.

Caveat against that MODIFY worth flagging: a 2025 Cell Reports paper reports a non-LPS lipid
ligand PMID:41264415. If that holds,
the generic lipid binding parent is independently informative rather than redundant.

3. Contested question B — stoichiometry and the "receptor" abstraction

Two 2026 structural papers agree that there is no clean 1:1 receptor–ligand complex.
PMID:41477831 — folding-on-binding, then
polymerization; and
PMID:42546204. The protease domain is monomeric until assembly
PMID:42546204.

This does not change any GO term, but it does mean GO:0160074 non-canonical inflammasome complex should be read as a heterogeneous assembly, not a defined stoichiometric complex.
Note that the PNAS NMR work is on caspase-11 CARD, not caspase-4 (see §5).

4. Contested question C — what does CASP4 cleave?

New substrate claim, 2026:
PMID:42044191, with a
direct biochemical demonstration using catalytically dead substrates and site mutants
PMID:42044191, feeding into GSDME
PMID:42044191 and concluding
PMID:42044191.

The same paper demotes the textbook CASP4→GSDMD axis in cells:
PMID:42044191,
which they say is consistent with their own earlier work
PMID:42044191.

My reading. This is a relative flux claim in THP-1 cells, not a claim that CASP4 cannot
cleave GSDMD — the same paper's own in vitro control shows it does
PMID:42044191 (verbatim text checked in cached full
text). Direct CASP4→GSDMD cleavage is established by reconstitution and structure
(PMID:26375003, PMID:32109412) and those remain unchallenged as biochemistry. So
GO:0004197 and the GSDMD-maturation annotations stand. What is genuinely unresolved is the
quantitative division of labour in intact human macrophages, and whether CASP3/CASP7 should
be curated as CASP4 substrates on the strength of one study. I did not add a NEW
annotation for CASP3/7 activation — one paper, and it is the exact point in dispute. It is
recorded in suggested_questions and in the reason on GO:0006915.

A second, older substrate dispute worth flagging: UniProt asserts CASP4 "does not directly
process IL1B (PubMed:7743998, PubMed:7797510, PubMed:7797592)", whereas
PMID:37558421.
Both positions are in the record; no GO term currently turns on it.

5. Mouse Casp11 is not a 1:1 ortholog — which annotations depend on it?

Mouse has a single Casp4 gene (MGI:107700, protein commonly called caspase-11); human has
the paralog pair CASP4/CASP5. Functional divergence is documented in the primary literature:
PMID:37993714, and
PMID:37993712 restricts pro-IL-18 cleavage to mammals "except rodents". PMID:42044191 finds
CASP11 processes CASP3/7 only "to a lesser extent". Conversely, cross-species complementation
does work for the core LPS→pyroptosis axis
PMID:25119034.

Annotations whose support runs through mouse Casp4/11 (checked against the PAINT file
interpro/panther/PTHR47901/PTHR47901-paint.tsv):

term	IBD node	seeds
GO:0070269 pyroptotic inflammatory response	PTN002573059	UniProtKB:P49662 (CASP4 itself), MGI:MGI:107700, zebrafish caspa/caspb
GO:0050729 positive reg. of inflammatory response	PTN002573059	UniProtKB:P49662 (itself), MGI:MGI:107700
GO:0072558 NLRP1 inflammasome complex	PTN002573059	UniProtKB:P51878 (CASP5), zebrafish caspa/caspb
GO:0005829 cytosol	PTN008306143	includes P49662 itself, P51878, MGI:MGI:107700
GO:0043525 positive reg. of neuron apoptotic process	PTN000047947	RGD:2275 Casp3, RGD:620945 Casp8, RGD:69274 Casp2, RGD:70967 Casp6, MGI:MGI:1277950 Casp9, WB:WBGene00000417 ced-3

Donor identities resolved via mygene.info; P51878 confirmed as CASP5 via UniProt REST.

The first three rows include CASP4's own experimental annotation among the IBD seeds,
which per project guidance is the expected, non-circular marker that the node is
experimentally grounded on the target itself. Those are ACCEPTed. The last two are the
problematic ones:

GO:0072558 NLRP1 inflammasome complex — the GO definition itself names only caspase-1
and caspase-5 ("An inflammasome complex that consists of two components, NLRP1 (NALP1) and
caspase-1 or caspase-5"). The IBD was seeded by CASP5 plus two zebrafish caspases; there is
no experimental placement of human CASP4 in an NLRP1 inflammasome. Marked
MARK_AS_OVER_ANNOTATED with a structured propagation_review (paralog transfer + complex
mismatch). CASP4's complex is GO:0160074.
GO:0043525 positive regulation of neuron apoptotic process — seeded exclusively by
apoptotic caspases (Casp2/3/6/8/9, ced-3) at the deep pan-caspase node PTN000047947, and
propagated to an LPS-sensing inflammatory caspase. Marked MARK_AS_OVER_ANNOTATED with a
structured propagation_review.

6. Other calls worth recording

GO:0005576 extracellular region (IEA + EXP, PMID:22246630) looked suspicious but is
grounded: PMID:22246630 and UniProt records "Released in the extracellular milieu by keratinocytes
following UVB irradiation". KEEP_AS_NON_CORE, not removed.
ER/mitochondrion localisation and ER-stress apoptosis come from a coherent older literature
PMID:15123740 and
PMID:23661706.
All KEEP_AS_NON_CORE — real, reproduced, but a separate arm from LPS sensing.
GO:2000494 positive regulation of interleukin-18-mediated signaling pathway (IDA,
PMID:33377178) is the wrong abstraction: CASP4 makes mature IL-18, it does not act inside
the IL-18 receptor signalling pathway. MODIFY → GO:0032741 positive regulation of interleukin-18 production.
GO:1903265 positive regulation of tumor necrosis factor-mediated signaling pathway (IDA,
PMID:16920334) comes from a HeLa overexpression study whose actual finding is a CARD–CARD
interaction PMID:16920334. MARK_AS_OVER_ANNOTATED.
Three bare GO:0005515 protein binding IPI rows (BioPlex PMID:33961781 with Q17R89;
cathepsin G PMID:29077095; SERPINB1 PMID:30692621) → MARK_AS_OVER_ANNOTATED per project
guidance. The SERPINB1 and cathepsin G interactions are real and mechanistically meaningful
PMID:30692621 — the objection is to the uninformative term, not the experiment.
GO:0032991 protein-containing complex (IDA) is likewise uninformative →
MARK_AS_OVER_ANNOTATED.
GO:0061702 canonical inflammasome complex is kept non-core: CASP4 does feed NLRP3/NLRP6
canonical inflammasomes PMID:26508369
PMID:33377178, but its own complex is the non-canonical one.

7. What I could not resolve

Whether the ligand of CASP4 should be curated as an LPS molecule or an LPS-membrane
surface of defined curvature. Kept as lipopolysaccharide binding; flagged.
Whether CASP3/CASP7 are genuine physiological CASP4 substrates. One paper, direct
biochemistry, but it is the disputed point; no NEW annotation added.
Whether CASP4 or CASP1 carries most GSDMD cleavage in human macrophages. Does not change
the MF annotation but would change how a GO-CAM should wire the pathway.
Whether CASP4 directly processes pro-IL-1β (PMID:37558421 yes; UniProt/1995 papers no).