CASP3 (Caspase-3) review notes

UniProt: P42574 (CASP3_HUMAN), 277 aa precursor. EC 3.4.22.56. Peptidase C14A family.
Synonyms: CPP32, Yama, apopain, SCA-1. Heterotetramer of two p17 (large) + p12 (small) subunits.

Core identity / molecular function

Substrate cleavage / downstream demolition (all DEPEND on the same MF, mostly downstream processes)

Localization

Regulation

Curation strategy

Exact supporting quotes gathered (for supported_by)

Falcon deep research integration (2026-06-21)

Falcon (Edison) deep-research report is in CASP3-deep-research-falcon.md; it strongly corroborates the existing
review/annotations (executioner cysteine-aspartate protease, EC 3.4.22.56, DxxD specificity, CASP8/9 activation,
PARP1/ICAD/GSDME substrates, cytoplasmic+nuclear localization, non-apoptotic differentiation roles, IFN restraint),
adding no contradictions but a few refinements. All Falcon-sourced citations below are not yet independently verified
against full text.

New / refined points (beyond existing notes):
- Refined cleavage motif: native-lysate subtiligase N-terminomics gives a stronger CASP3 consensus DEVD↓(G/S/A) at
P4-P1↓P1' (not just generic DxxD↓), with cleavage also gated by local structural accessibility. [Araya et al.,
ACS Chem Biol 2021, doi:10.1021/acschembio.1c00456; Soni & Hardy, Biochemistry 2021, doi:10.1021/acs.biochem.1c00459]
(not yet independently verified against full text)
- Scale of substrate repertoire quantified: ~906 putative substrates / 1126 cleavage sites in native lysates, of which
577 sites and 257 substrates were novel vs the apoptosis DegraBase; substrates ~49% cytoplasmic and ~48% nuclear.
[Araya et al., ACS Chem Biol 2021, doi:10.1021/acschembio.1c00456] (not yet independently verified against full text)
- CASP3 vs CASP7 non-redundancy: human GSDME is cleaved efficiently by CASP3 but not by CASP7 (attributed to CASP7 p10
subunit / prime-side recognition differences), supporting CASP3 as the dominant/most proteolytically proficient
executioner. [Blais & Denault, Biosci Rep 2026, doi:10.1042/bsr20254030; Mustafa et al., Cells 2024,
doi:10.3390/cells13221838] (not yet independently verified against full text)
- New substrate: CAD (carbamoyl-phosphate synthetase 2/aspartate transcarbamylase/dihydroorotase, CAD enzyme of
pyrimidine synthesis) cleaved at Asp1371 prior to degradation, linking CASP3 to pyrimidine synthesis control and
chemosensitivity. [Tannous et al., Bioconjug Chem 2025, doi:10.1021/acs.bioconjchem.5c00151] (not yet independently
verified against full text) — note potential ambiguity: report's "CAD" elsewhere also denotes caspase-activated DNase.
- Non-apoptotic cytoprotective autophagy / DDR: CASP3+CASP7 promote starvation- or proteasome-inhibition-induced
cytoprotective autophagy and the DNA-damage response in breast cancer cells; loss reduces LC3B/ATG7 and H2AX
phosphorylation, and is synthetic-lethal with BRCA1 deficiency. [Samarasekera et al., PLoS Biol 2025,
doi:10.1371/journal.pbio.3003034] (not yet independently verified against full text) — relevant to existing
GO:0016241 regulation of macroautophagy annotation.
- IFN restraint mechanism specified: CASP3/7 prevent MOMP-released mtRNA from triggering the MDA5/MAVS/IRF3 type I IFN
pathway, keeping apoptosis immunologically silent. [Killarney et al., Nat Commun 2023, doi:10.1038/s41467-023-37146-z]
(not yet independently verified against full text) — consistent with existing GO:0001818 negative regulation of
cytokine production / cGAS-MAVS-IRF3 cleavage notes (PMID:30878284).
- Neural stem cell differentiation: non-apoptotic CASP3/7 events bias adult NSCs toward direct (division-independent)
neuronal conversion, requiring transcription factor Atf3 (zebrafish tracer). [Rosa et al., Development 2024,
doi:10.1242/dev.204381] (not yet independently verified against full text) — supports existing neuron-differentiation
/ non-core developmental annotations.
- Non-apoptotic synaptic role + context-dependent pro-oncogenic role: CASP3 drives synaptic pruning / dendritic-spine
loss / LTD without death (Parkinson model), and via EndoG->nucleus and Src-STAT3 can promote oncogenic transformation.
[Fieblinger et al., Int J Mol Sci 2022, doi:10.3390/ijms23105470] (not yet independently verified against full text) —
consistent with existing synapse-pruning (GO:0098883) non-core annotations.

Discrepancies / annotations to revisit:
- No direct contradictions with the existing review were found; Falcon is consistent with the curated picture.
- The execution/substrate-cleavage themes (autophagy regulation, IFN/cytokine restraint, NSC and synaptic roles) are
already captured as ACCEPT or KEEP_AS_NON_CORE, so no action implied for those. The only candidate for a possible
NEW supporting annotation is the CASP3-vs-CASP7 substrate-discrimination point, but this is a comparative-biochemistry
observation rather than a new GO term for CASP3 itself; no annotation change recommended from Falcon alone.

GO:0005515 re-review (2026-09-28)

2026-10-25 comparative apoptotic-execution review

Systematically examined all 199 rows, preserving every imported annotation payload. Decisions judge the biological claim, not the evidence code: independently established cysteine endopeptidase activity, cytosolic/nuclear localization and apoptotic execution are retained/refined even when one original citation is not sufficient alone. No experimental claim is rejected merely because the cached abstract foregrounds a different protein. Bare protein-binding removals mean uninformative annotation, not false interaction.

Species and paralogs: human CASP3/CASP7 have distinct natural-substrate profiles (PMID:18723680). Mouse CASP7 can cleave ICAD efficiently and contributes to strain-dependent compensation for Casp3 loss (PMID:15525783). The 129X1/SvJ versus C57BL/6J comparison in PMID:12152782 must not be conflated with the 129S1/SvImJ comparison in PMID:15525783. These results do not establish universal redundancy or a single obligatory executioner across tissues.

CASP6: full PMID:32029622 establishes direct CASP3/CASP7 cleavage of procaspase-6 in vitro (Fig. 6A), inhibitor-sensitive procaspase-6 processing in primary hepatocytes (Fig. 6B), and a CASP6–BID mitochondrial feedback mechanism; Fig. 6H uses HepG2 cells. It does not justify a universal initiator-caspase activation route. Full PMID:32298652 and public primary Figure 6A identify mouse CASP6/RIPK3 constructs despite the HEK293T host; native infection experiments also use mouse cells/mice. Human prior NEW adaptor annotation is retired/UNDECIDED; conserved contextual human immune claims remain explicitly comparative. Mouse IAV findings do not imply a requirement for all inflammasome triggers.

CASP7: full PMID:35705808 establishes mouse SMPD1 processing and membrane-pore repair. CASP7 performs proteolytic maturation; SMPD1 performs ceramide-producing lipid hydrolysis. Repair delays lysis in the tested settings, so this mechanism is not universally pro-lytic. PMID:22464733 distinguishes PARP1 cleavage from PARP1 poly-ADP-ribosylation activity.

Evidence handling: checked existing primary caches and retained unresolved precise cellular/stimulus claims as UNDECIDED where evidence could not establish the claim. 0 pre-existing nonverbatim deep-research quotations were removed from the optional quotation field; source citations and review rationales remain, with direct primary evidence preferred for core functions. Provider reports remain immutable. Existing reports were used where present; missing Falcon reports were attempted with perplexity-lite fallback, and failures are recorded in .cache/module-audit-logs/apoptosis-caspase-*-research.log. No missing report was fabricated. Sources were fetched through normal helpers only.

Verified UniProt PANTHER placement: PTHR10454 CASPASE; CASP3/Casp3 SF198 CASPASE-3, CASP6/Casp6 SF206 CASPASE-6, CASP7/Casp7 SF31 CASPASE-7. These are membership checks, not assertions of identical physiological deployment.

The annotation-reviewer consultation confirmed that true same-target catalytic/location claims should not be removed for citation weakness, and that substrate cleavage is not the same as the substrate's enzymatic reaction. Notes are comparative; only target-specific claims are asserted in the review.

Primary-support follow-up: 8 retained annotation rows received claim-specific primary/curated support; 26 precise contextual claims without adequate inspected support were explicitly set UNDECIDED. Provider citations remain in the bibliography but are no longer the sole asserted evidence for these retained claims.

PR5558 source-specific transfer audit (2026-10-26)

Reassessed 25 previously templated UNDECIDED rows. The full per-row provenance, primary-study scope, access status and decision are recorded in genes/human/CASP3/CASP3-source-analysis/transfer-evidence-audit.tsv. Seed WITH/FROM and live QuickGO donor records were used only to trace publications; they are not evidence for a function, and no relationship qualifier was interpreted. All biological quotations come from the normally cached primary papers. No original row field, row count or core function was changed.

Response terms can include changes in enzyme production/gene expression, so a drug-dependent abundance change is not automatically rejected for lacking a gene knockout. This does not make CASP3 a drug receptor or prove apoptotic activation. Conversely, an expression marker accompanying a behavioral test does not establish swimming locomotion. Cellular-component claims were assessed from donor localization: neuronal cell bodies from rat PrP106-126 injury, and postsynaptic/glutamatergic compartments from the mouse spine study. Neuronal differentiation has independent neurosphere inhibition support alongside the rat observational source and existing PAINT inference; no ancestral reconstruction or universal neuronal requirement was invented.

Nine claims remain unresolved across the two reviews, each with a specific cited assay/access limit: axon bundling, hippocampal anatomical development, glucose or nicotine stimulus specificity, exact anterior neural-fold fusion, and steady-state T-cell population homeostasis. Full PMID:21515572 establishes the Casp3 hith2 encephalocele phenotype but does not by itself identify anterior-fold fusion; the Scrib crn2 phenotype in the same paper is kept separate without accusing the curator of a gene mix-up. Full PMID:18701083 supports activated-T-cell restraint during anergy, while the original PMID:12970760 abstract foregrounds B cells and does not expose every T-cell assay. No negative verdict is inferred from a title or missing abstract detail.

Normal PMID force-fetches were used to recover available full text; the first21515572 fetch failed, and a normal retry obtained full text. PMID16103108 and18701083 were resolved through EuropePMC and fetched normally. Existing provider reports were not rewritten or fabricated. The normal caches still lack full text for several donor studies, explicitly recorded in publication-access.json.

PR5558 residual uncertainty audit (2026-10-26)

Reassessed 6 additional rows after the first37-row pass. All remaining uncertainty reasons in both human and mouse reviews were inspected, including binding, swimming, cytokine/signaling and source-specific experimental rows rather than only the repeated stimulus template. All original annotation payloads, flags, row counts and cores remain unchanged.

The rat IEA stimulus rows were matched by term, evidence, reference and exact WITH/FROM provenance before reusing the existing primary assessments. Two rat ISO rows use RGD:2275 for the same donor protein. Glucose/nicotine retain the original source-specific access boundaries; expression-only responses do not require direct sensing. Staurosporine is supported by human PMID:22253444 and native mouse full PMID:16469926. The latter also directly resolves UV-response genetics; failure of Casp3 loss to abolish death in neural precursors (abstract-only PMID:11092819) does not refute participation in a response in other cells. No relation qualifier was used as biological evidence.

Normal refreshes of PMID:17518537,11092819,16469926,23954828,15246877 succeeded; only16469926 gained full text. PMID:22949508 was newly normally fetched in full and supplies native mouse Casp3-knockout/RasGAP-cleavage-resistant evidence for an anti-apoptotic protective response. Historical human GO:0043066 and GO:0035655 donor assertions were not recovered by current exact-term queries; annotation-state absence is not counterevidence. The former is resolved by direct mouse biology, while IL-18 cleavage/inactivation (PMID:9334240) does not settle receptor-initiated signaling. The three rat DISC-associated binding classes are each bounded by the exact missing assay in abstract-only PMID:17518537. All generic apoptotic-effector quotes were removed from UNDECIDED rows.

The committed source audit now uses authored YAML with a deterministic TSV export, following the explicit CLAUDE.md structured-data rule. The established source-analysis folder remains because these are source/provenance checks, not sequence bioinformatics.

Row (zero-based) GO term Action Source-specific assessment
39 GO:0002020 protease binding UNDECIDED Provenance: GO_REF:0000107 from UniProtKB:P55213, traced to PMID:17518537 in adult female Fischer rat spinal cord after injury. The rat donor annotation identifies a protease partner (RGD:620945, Casp8). The abstract places caspase-8 and caspase-3 in the SCI-associated DISC but does not expose a pair-specific binding assay that separates protease-substrate encounter from a stable interaction. Normal force retrieval remains abstract-only. The exact experimental distinction cannot be settled from that abstract, and no binding absence, wrong-protein error, or target-species loss is inferred. The source association is genuine; this uncertainty is about the molecular claim and its transfer, not the general apoptotic role of CASP3.
43 GO:0005123 death receptor binding UNDECIDED Provenance: GO_REF:0000107 from UniProtKB:P55213, traced to PMID:17518537 in adult female Fischer rat spinal cord after injury. The rat donor evidence concerns Fas/CD95 association after spinal cord injury. Its abstract describes a DISC containing Fas, FADD, caspase-8, cFLIP and caspase-3, but does not distinguish a caspase-3/Fas binding interaction from association bridged through the complex. Normal force retrieval remains abstract-only. The exact experimental distinction cannot be settled from that abstract, and no binding absence, wrong-protein error, or target-species loss is inferred. The source association is genuine; this uncertainty is about the molecular claim and its transfer, not the general apoptotic role of CASP3.
66 GO:0036269 swimming behavior MARK_AS_OVER_ANNOTATED Provenance trace: GO_REF:0000107 via UniProtKB:P55213 / ensembl:ENSRNOP00000014096. The complete rat source measures Morris-water-maze performance after splenectomy and IGF1 treatment while measuring hippocampal caspase-3 as an expression marker. It does not manipulate CASP3 to establish a locomotor swimming role; moreover, the assay is interpreted as cognitive performance, not a CASP3-dependent swimming motor program. GO:0036269 specifically denotes the locomotory response of swimming. Transferring this assay-associated marker into a human swimming-behavior function overstates the source context. This does not assert that caspase-3 cannot influence neurons, cognition or behavior in other experiments.
73 GO:0044877 protein-containing complex binding UNDECIDED Provenance: GO_REF:0000107 from UniProtKB:P55213, traced to PMID:17518537 in adult female Fischer rat spinal cord after injury. The rat donor evidence describes recruitment into a Fas-associated DISC. Its abstract does not resolve whether the measured event establishes binding of caspase-3 to the assembled macromolecular complex, versus membership or co-recruitment inferred from its component associations. Normal force retrieval remains abstract-only. The exact experimental distinction cannot be settled from that abstract, and no binding absence, wrong-protein error, or target-species loss is inferred. The source association is genuine; this uncertainty is about the molecular claim and its transfer, not the general apoptotic role of CASP3.
123 GO:0005515 protein binding UNDECIDED Normal force retrieval of PMID:15246877 remains abstract-only. It describes S-nitrosation of thioredoxin and dissociation/activation of ASK1 in the NO/superoxide system, but does not expose the CASP3 interaction experiment underlying this IPI row. The original full text may contain that experiment; its absence from the abstract is not evidence of wrong-gene attribution. Preserve uncertainty rather than remove a binding assertion whose precise evidence cannot be adjudicated.
135 GO:0048011 neurotrophin TRK receptor signaling pathway UNDECIDED PMID:23954828 separates mouse DRG expression/colocalization from HEK293T experiments in which TrkC induces apoptosis, NT-3 protects, and Dok5 knockdown changes caspase-3 activity. The normal force retrieval remains abstract-only. Those results establish a caspase readout downstream of the survival/death manipulation, but do not expose the experiment distinguishing a CASP3-performed receptor-signaling step from downstream apoptotic execution. No absence of CASP3 function or false curator assignment is inferred; the precise pathway participation and target-species transfer remain unresolved.

PR5558 remaining peripheral/core and execution-source audit (2026-10-26)

The 16-member panel baseline at bb9c384a3960307c636d030a30a5ab17879582c4 contains 1,608 rows; all original payloads and raw UniProt/GOA hashes are preserved. This follow-up changes only CASP3/Casp3 reviews and their evidence documentation, without new annotation rows or module-graph changes. The audit source directory preserves official GO definitions/ancestry and failed publisher-access attempts.

Positive apoptotic regulation is retained as ACCEPT: direct substrate cleavage and propagation of the caspase cascade (normally recovered full 18723680), with native mouse Casp3/Casp7 mitochondrial feedback (full 16469926), support a true effector/amplifier role. GO:0097194 execution phase is not an is_a/part_of descendant of GO:0043065 positive regulation, so it is not a valid ancestor-refinement argument. Intrinsic signaling ends when execution is triggered according to its official definition; the retained signaling assertion is therefore grounded in feedback on early mitochondrial events, not merely downstream execution. Double-knockout phenotypes are not attributed uniquely to Casp3.

Full 36002459 directly tests Casp3 siRNA in native mouse macrophages and Casp3-null mice, as well as human CASP3 biochemical and resistant-substrate experiments. Both species support cytokine suppression as a mechanistic core role. The 30878284 abstract explicitly distinguishes additional mouse Casp7 participation in cGAS inactivation; it does not support identical paralog dependency. Human core language now retains that limit.

The GSDME mechanism remains conditional. The broader pyroptotic inflammatory-response claims remain non-core context; the specific positive-regulatory cleavage activity is accepted in both species and represented as a conditional mouse core alongside the human core. Full 33852854 separates sublytic IL-1beta release from lysis and uses human GSDME to rescue mouse iBMMs that have little endogenous GSDME. That experiment is not rewritten as a native mouse-GSDME assay. Human epithelial CASP3/GSDME knockout results in full 35594856 have a different cell context. The 28459430 abstract reports human primary-cell responses and Gsdme-null mouse protection, not a Casp3 knockout in every context.

Protein-localization claims retain the specific DSG2/JUP context from 17559062: human HaCaT/HT-29 cells, released cytosolic DSG2 fragment, lost surface staining and z-DEVD-fmk protection. The mouse claim remains an explicit orthology inference, not an invented mouse imaging experiment or transport-carrier activity.

All apoptotic-process refinements now use direct execution/source evidence rather than C14A-family membership. Inspection found nine mouse rows (including the 17901126 row omitted from the review's count of eight) and five human rows with this support problem; all fourteen were addressed. Original source context remains explicit when a paper measures active Casp3 as a readout of another perturbation.