Falcon deep research completed for human AFG3L2/Q9Y4W6 and was used as a synthesis check against the cached GOA-linked literature. The report consistently identifies AFG3L2 as the human mitochondrial inner membrane m-AAA protease subunit, not a soluble matrix protease.
AFG3L2 is an m-AAA protease subunit anchored in the mitochondrial inner membrane, with the catalytic AAA+ ATPase and zinc metalloprotease modules exposed toward the matrix side. The topology point is important for PN projection: the protein acts on matrix-facing substrates but is itself an inner membrane protein PMID:31327635. Older biochemical work also recovered paraplegin and AFG3L2 in the membrane fraction PMID:14623864.
The core activity is ATP-dependent substrate unfolding/translocation coupled to metalloprotease cleavage. Structural work reports that AFG3L2 has features required for "ATP-dependent translocation to unfold and degrade targeted proteins" PMID:31327635, and biochemical work describes human AFG3L2 as an "ATP-fueled degradation" protease at the matrix face of the inner membrane PMID:29932645. ATP hydrolysis is a real catalytic activity, not just a binding annotation PMID:19748354.
The biological process center is mitochondrial substrate degradation and processing. AFG3L2 is required for mitochondrial protein quality control of newly synthesized or misassembled inner-membrane proteins PMID:26504172 and resolves MT-ATP6 insertion defects PMID:34718584. It also processes selected substrates such as MRPL32 PMID:29932645.
Several substrate-specific consequences are well supported but should be treated as non-core relative to the m-AAA protease function. AFG3L2/SPG7 degrade unassembled EMRE/SMDT1 to regulate MCU complex assembly and calcium import [PMID:27642048 "the m-AAA protease degrades non-assembled EMRE and ensures efficient assembly of gatekeeper subunits with MCU"; PMID:28396416 "mitochondrial mAAA proteases AFG3L2 and SPG7 rapidly degrade unassembled EMRE using the energy of ATP hydrolysis"]. AFG3L2 also degrades SLC25A39 in a glutathione-sensitive feedback circuit [PMID:37917749 "Under physiological conditions, SLC25A39 is rapidly degraded by mitochondrial protease AFG3L2."; PMID:38157846 "mitochondrial m-AAA protease AFG3L2 is responsible for degrading SLC25A39 through the matrix loop 1"].
The PN projection proposed two AFG3L2 candidate additions from Mitochondrial proteostasis|Organelle-specific protein degradation|Matrix protease:
GO:0035694 mitochondrial protein catabolic processGO:0005759 mitochondrial matrixI accepted GO:0035694 mitochondrial protein catabolic process as a conservative NEW annotation. Existing GOA already has generic GO:0030163 protein catabolic process and generic GO:0006508 proteolysis; the literature supports a mitochondrial-specific degradation process, and the PN report itself states that "GO mitochondrial protein catabolic process is the conservative shared target."
I did not add GO:0005759 mitochondrial matrix. The PN class label "Matrix protease" reflects the side of the inner membrane where the catalytic domains face and where many substrates are exposed. It does not override the stronger cellular component evidence that AFG3L2 is an integral mitochondrial inner membrane protein. The safer cellular component remains GO:0005743 mitochondrial inner membrane, including is_active_in when available.
Generic proteolysis and protein catabolic process annotations were marked for replacement by GO:0035694 mitochondrial protein catabolic process. Broad membrane or mitochondrion cellular component annotations were marked for replacement by GO:0005743 mitochondrial inner membrane.
The protein binding IPI rows were treated as over-annotations. The reported interactions are real, but the generic MF term does not communicate AFG3L2 function. Where appropriate, the informative biology is captured by m-AAA complex, substrate degradation, mitochondrial calcium import regulation, glutathione-response biology, or mitochondrial protein quality control.
This audit supersedes the earlier review judgments where explicitly described below; the historical notes, provider report and generated PN notes remain intact. The approved HGNC symbol is AFG3L2 (HGNC:315), with previous symbol SCA28 and alias SPAX5 in the archived official HGNC subset. No historical-symbol gene directories were found. Read-only GitHub checks found no open AFG3L2/SCA28/SPAX5 PR, and all local gene files matched main commit 795b693f5711c625401d03a755fe937260ae0ac0 before editing. The starting review contained 77 machine-seeded assertions plus one prior NEW proposal. All 77 original source objects, including terms, references, evidence types, qualifiers and supporting entities, are preserved. All 39 original reference identifiers/titles remain exact.
Research and source access:
just fetch-gene-pmids human AFG3L2. Both research commands failed while installing deep-research-client from PyPI because DNS resolution failed, before either provider ran. No report was fabricated or overwritten. The existing genuine Falcon report remains historical context, independently checked against primary evidence.fetch-pmid 19656850 attempt failed DNS and produced no file. The review remains DRAFT until that missing cache is obtained. Primary verification and local cache availability are separate facts.AFG3L2 is a membrane-anchored, matrix-facing protease rather than a soluble matrix protein. PMID:14623864 directly combines mitoplast fractionation, alkaline extraction and reciprocal AFG3L2–SPG7 coimmunoprecipitation in human cells: "Paraplegin and AFG3L2 were recovered from the membrane fraction, indicating that both are integral proteins of the mitochondrial inner membrane". The original broad mitochondrial immunofluorescence assertion in PMID:10395799 remains ACCEPT at its own resolution. HPA/immunofluorescence, mitochondrial HTP, the PINK1 study and InterPro membrane rows also retain their supported broad compartments; separate annotations establish the finer inner-membrane location.
PMID:19748354 measures coordinated ATP hydrolysis in purified human AFG3L2 homo- and SPG7-containing hetero-oligomers expressed in yeast. Walker A and Walker B effects differ. Its MrpL32/Ccp1 processing experiments concern yeast substrates, not native human MRPL32. PMID:29932645 directly tests engineered soluble coiled-coil-hexamerized human AFG3L2, protein degradation, short-peptide cleavage and a recombinant human MRPL32 precursor lacking its first 30 residues. The processed product is characterized by mass spectrometry. Protein unfolding/degradation is ATP-fueled; accessible short-peptide cleavage can occur without ATP. PMID:31327635 resolves a membrane-truncated human catalytic-core homohexamer with ATPase/protease-inactivating mutations, while separate active engineered-enzyme assays establish ATPase and cleavage functions. Inactive trapped structural density is not itself a catalytic turnover assay.
These data support one integrated core activity, ATP-dependent peptidase activity, with ATP-driven unfolding/translocation and zinc-dependent peptide cleavage. Mitochondrial protein quality control, catabolism and productive precursor processing are retained as distinct aspects of that function. This replaces the earlier two core units that separated the motor and cleavage domains of the same machine. No new annotation is proposed.
Live ontology checks on 2026-09-27 confirmed that GO:0034982 is a child of protein processing, GO:0016485, whose definition concerns proteolytic maturation to functional capacity. Consequently the source-specific SLC25A45 processing row is changed to mitochondrial protein catabolism: PMID:41075794 Figure 2 and Figure S6 measure stabilization and slowed turnover after AFG3L2 depletion, not production of a mature SLC25A45 fragment. This does not challenge AFG3L2's separately demonstrated precursor-maturation capacity. In the 2019 structural-source processing row, acceptance explicitly relies on independent 2018 human precursor biochemistry as well as curator judgment, rather than relabeling model-substrate destruction as maturation.
The previous NEW GO:0035694 proposal is withdrawn because several retained MODIFY replacements already cover mitochondrial protein catabolism. Its source assertion was a pre-existing proposal rather than a seeded GOA row. The historical Proteostasis projection files remain unchanged; they are not primary experimental evidence.
The MGI Q8JZQ2 sequence record identifies the ISS/Compara donor as mouse Afg3l2. The comparative GO snapshot, generated 2023-03-10, links mouse autoprocessing and protein processing to PMID:19656850. Primary PubMed confirms Autocatalytic processing of m-AAA protease subunits in mitochondria., DOI 10.1091/mbc.E09-03-0218. Indexed original PMC2754935 Results and Figures 4–5 were read on 2026-09-27: mouse Afg3l2 undergoes MPP-dependent intermediate formation followed by maturation requiring existing Afg3l1/Afg3l2, examined in yeast and mouse mitochondrial import experiments. This supports a real donor function, not merely a predicted precursor. Conserved human architecture and the curated human processing record support transfer as non-core biogenesis. A full-length human self-cleavage assay was not recovered, and the mouse Afg3l1 contribution is not transferred to human as an active third subunit.
PAINT reviews retain only ancestral PTN nodes as proximate source entities. The human target's presence among descendant experiments is legitimate grounding; donor counts are not used to weaken support. InterPro mappings are assessed against actual human enzyme/topology data. ARBA identifiers are retained, but unrecovered rule internals are explicitly UNRESOLVED. No phylogeny, alignment or current Ensembl tree was reconstructed. These limits do not negate direct human evidence supporting the biochemical judgments.
PMID:26504172 Figures 8–9 establish nascent-chain proteostasis and actinonin responses. Five-day AFG3L2 depletion did not reduce large mitoribosomal-subunit assembly or monosome formation. PMID:30683687 shows delayed ribosomal/membrane consequences and suppression by mitochondrial-translation inhibition, separating primary nascent-chain stress from later phenotypes. PMID:34718584 shows that combined OXA1L/AFG3L2 depletion prevents rapid degradation of newly synthesized MT-ATP6 and MT-CO2 observed after OXA1L depletion. These support direct quality-control participation, not a universal claim that every human phenotype follows MRPL32 maturation failure. OMA1 performs stress-induced OPA1 cleavage; AFG3L2 is not assigned that downstream reaction.
The original PMID:27642048 abstract directly reports mouse neuronal interactome, MAIP1 complex and EMRE turnover context. Full source recovery failed: the publisher PDF was inaccessible, the university archive marked its copy restricted, and ResearchGate offered a request. Exact human pair/construct details and the axonogenesis experiment therefore remain unresolved. The axonogenesis IMP row is UNDECIDED; a general axonal-development sentence in the later structural paper does not replace this original experiment. No wrong-species inference is made from the abstract.
Primary full PMC5410796 for PMID:28396416 was read externally on 2026-09-27, including Figures 1–5. AFG3L2/SPG7 depletion slows degradation of unassembled EMRE in HEK293 cells; independent shRNAs and isolated-mitochondrial ATP-dependence assays reinforce the mechanism. Protease-resistant EMRE deregulates uptake when MICU gatekeeping is limiting, but overexpressed resistant EMRE alone did not yield detectable low-calcium uptake with sufficient MICU1. Source-specific calcium regulation/homeostasis remains NON_CORE. The import row is refined to regulation on mechanistic grounds, not because only channels can participate in transport.
Cached GO-CAM 65c57c3400001115 explicitly models AFG3L2 metalloendopeptidase activity with EMRE as input and negative causal influence on the EMRE-dependent transport machinery; MCU supplies channel activity. Model 65c57c3400001018 similarly separates AFG3L2 proteolysis from SLC25A39 glutathione import/iron-sulfur binding. These support the role distinctions. No NEW process term is needed.
PMID:37917749 and PMID:38157846 establish AFG3L2-dependent SLC25A39 turnover, substrate-loop dependence and responses to GSH/iron status. SPG7 is dispensable for the SLC25A39 control tested in the former paper. The substrate bears the metabolite-responsive features; AFG3L2 supplies proteolysis. PMID:35912435 demonstrates TMBIM5 association/inhibition of AFG3L2 and AFG3L2 contributions to TMBIM5 turnover under hyperpolarization, while allowing other proteases. Neither TMBIM5 calcium/proton exchange nor SLC25A39 transport is reassigned to AFG3L2.
For PMID:22354088, externally indexed primary PMC3321149 Results/Figures 1–3 were recovered on 2026-09-27 after direct opening challenged. AFG3L2 silencing changes PINK1 cleavage-intermediate/product abundance under membrane-potential controls. The authors infer direct participation, but no purified AFG3L2–PINK1 cleavage reaction or exact AFG3L2 cleavage site is established by those assays. MPP has a distinct upstream processing/import role. The metalloendopeptidase and membrane-protein-proteolysis annotations are retained with those bounds. OMA1 maturation in PMID:29545505 is explicit in the cached abstract; unavailable full details are not invented.
All six generic protein-binding annotations are REMOVE under the repository's functional-informativeness policy. This is not a claim that the interactions are false. PMID:14623864 and PMID:26387735 directly support AFG3L2–SPG7 association; the latter's permeability-transition interpretation is separate from that interaction. PMID:35912435 directly supports TMBIM5 association. The original author manuscript for PMID:32814053 was re-read: repeated Y2H screening, pairwise retesting, dataset integration and selected validation preclude describing the work as one unvalidated interaction experiment. The exact AFG3L2–HTT supplementary pair was not recovered. In the two PMID:27642048 binding rows, exact human MAIP1/EMRE pair evidence remains uncertain; missing detail prevents selecting an informative mechanistic replacement, while removal of the generic term leaves no assertion of nonbinding.
All nine cached Reactome summaries were read. The PMPCA/PMPCB event is not relabeled as AFG3L2 transit-peptide cleavage. Matrix substrate access is distinguished from soluble matrix residence. The summary naming MRPL32 among degraded substrates is not used as proof that mature MRPL32 is completely destroyed; primary limited-precursor processing is stated explicitly.
The independent coordinator read all 77 judgments and the integrated core. The resulting refinements preserve broad source-level locations and distinguish generic-term removal from unresolved pair-level evidence. Validation and publication manifests record the final counts and the remaining PMID:19656850 cache gate.
This section supersedes the earlier statement that a human-protein self-maturation assay was not recovered, the NON_CORE autoprocessing decisions, and the earlier missing-cache status for PMID:19656850. The baseline is published PR head d81a4c9ba7a837234a0a5e3ce2e05b7b242c6832. All 77 original source objects and all 40 existing reference identities are unchanged.
PMID:30252181 was verified through official PubMed and its linked accepted manuscript. Methods, Results and Discussion were read. Figure 3C compares human AFG3L2 with the E575Q protease-inactive control and disease variants in yeast lacking endogenous m-AAA protease. Precursor accumulation supports self-maturation by human AFG3L2. Figure 3D separately supports AFG3L2-dependent paraplegin maturation; R468C retains complex association despite impaired processing. The Methods identify a Yta10(1–63)-human paraplegin(59–795)-HA targeting fusion, so this is not a direct assay of native human SPG7's MPP cleavage. The two existing autoprocessing assertions are now ACCEPT: AFG3L2 performs the cleavage in an integral enzyme-maturation reaction. This is not substrate-only participation. No NEW annotation is needed.
The core description and relevant mitochondrial-processing reviews now include self- and partner-maturation alongside the independently demonstrated recombinant human MRPL32 reaction. Yeast MrpL32/Ccp1 assays are not relabeled human substrates. Patient fibroblasts retain normal MRPL32 processing; altered OPA1 products do not establish direct AFG3L2 cleavage of OPA1. Native human-cell processing requirements remain appropriately bounded. A sibling's focused conceptual consultation agreed that direct self-cleavage can be core, while explicitly not claiming a new independent full-paper read.
The standard reference-recovery workflow produced publications/PMID_19656850.md, imported without modification after artifact and per-file hash verification. The record is abstract-only and its title exactly matches the retained reference. Its abstract supports mouse Afg3l2 autoprocessing and paraplegin maturation; earlier externally recovered Results remain separately scoped. It is retained because it grounds the original mouse donor, even with independent human-protein evidence. Provenance: source workflow run 36286975328, source head 5946477c8ac79ade0709264c775ea1262b108438, artifact 10920674630, transport run 36288441414; the exact receipt is tmp/verified-reference-records/local-import-receipt.json. Its bytes are included explicitly in this follow-up's publication manifest.
A normal fetch-pmid 30252181 attempt failed DNS with nodename nor servname provided, or not known, caching 0/1. No publication record or provider report was fabricated. The citation is primary-verified, but its machine cache remains missing; the review stays DRAFT. The new source's local full_text_unavailable: true flag does not deny the externally read manuscript. Existing provider artifacts remain unchanged. The notes-inclusive citation census now has 21 distinct PMIDs, with only PMID:30252181 missing.
The HTT generic-binding removal retains its functional-informativeness rationale; the unrelated introductory quotation was deleted. Broad membrane localization remains ACCEPT because membrane anchoring is core biology and is entailed by the finer inner-membrane location, rather than becoming non-core merely through GO granularity. The one integrated protease core retains explicit zinc-dependent endopeptide chemistry. An ontology question now asks how to represent both ATP-coupled translocation and metalloprotease cleavage without duplicating the same activity; short-peptide cleavage and ATP-driven protein degradation have different energetic requirements. The existing generic-binding removals do not deny interactions or pretend unavailable pair assays were read. The source-specific axonogenesis assertion remains UNDECIDED.
Only two action changes were made: the two autoprocessing rows move from KEEP_AS_NON_CORE to ACCEPT. Final counts are 49 ACCEPT, 13 MODIFY, 8 KEEP_AS_NON_CORE, 6 REMOVE and 1 UNDECIDED, with one integrated core and 41 reference assessments. Validation, rendering and exact source preservation are recorded in the follow-up manifest; the only unresolved publication cache is PMID:30252181.
The two seeded protein-autoprocessing assertions remain ACCEPT. AFG3L2 performs
the peptide-cleavage step itself; this is direct participation in its own
maturation, not an inference from being required for another enzyme’s work. The
mouse donor evidence and independent human-protein experiments remain separate:
PMID:30252181 expressed human AFG3L2 in m-AAA-deficient yeast, rather than native
human cells. No source field, annotation action or evidence code changes.
The integrated core now explicitly includes GO:0016540 protein autoprocessing.
The live GO autoprocessing record
and mitochondrial processing record
were checked on 2026-09-27: both are separate children of GO:0016485 protein
processing, and neither is an ancestor of the other. The added synthesis term
records the already-reviewed self-cleavage mechanism; it creates no NEW row.
Both annotation rows and the integrated core again cite the exact cached UniProt
PTM passage “autocatalytic proteolytic processing to generate the proteolytically”,
whose immediately following source line identifies the active mature form and
attributes it to PMID:30252181. This supplements the independently read primary
full-paper quotation with a mechanically checkable local source. The full-paper
quotation retains its explicit external-access field until a normal cache is
recovered and inspected. The published PR description will report the current
49 ACCEPT / 8 KEEP_AS_NON_CORE tally and distinguish the already closed mouse
PMID:19656850 cache gap from the human PMID:30252181 cache requirement.
Imported the exact normal-fetch abstract for PMID:30252181 from source recovery run 36289953066, artifact 10923045788. The record SHA-256 is f8b3d9d98413752ea2db983397fe897aaab911db7cf324b08739f2a534a86d73. The title and identifier match the reviewed study; the abstract supports the yeast functional analysis and patient OPA1/network phenotype. It contains no Figure 3, so the earlier accepted-manuscript assessment remains the source of construct-specific self-maturation details. full_text_unavailable: true accurately reflects this normal abstract cache. The exact cached UniProt quote remains the immediately checkable support attached to self-processing decisions and the integrated core. All 77 seeded source objects and action decisions are unchanged. This closes the last normal-cache gate; the source-specific processing-action and unused-provider advisories remain intentional.
The fresh main baseline was 30ad653fd840321991819c92c78f5127a9d3c011; all twelve gene/history files matched exactly, and canonical AFG3L2 and historical SCA28/SPAX5 PR searches were empty. HGNC:315 identifies approved AFG3L2, UniProt Q9Y4W6. The four required provider-linked publications below were recovered by the normal source-12 fetcher and imported unchanged after strict artifact, primary-identity and access checks. Their actual contents were read for this follow-up. This supersedes the earlier claim that recovery of the last annotation-linked abstract alone closed all citation requirements: the retained provider reports also needed this DOI-inclusive assessment.
| Reference | Actual access and interpretation |
|---|---|
| PMID:36301938 | Local abstract plus externally indexed original PMC10171965 inner-compartment section/Table 1. A review of mitochondrial structure and proteostasis, not new AFG3L2 or glaucoma-specific experiments. The local full-text-unavailable flag remains true. |
| PMID:37804316 | Local original Methods, Results and Discussion. Human patient fibroblasts and AFG3L2-reduced HEK293T cells are distinct from Afg3l2-null mice and mouse Purkinje cultures. Human DELE1-HA was overexpressed in HEK293T cells; endogenous cleavage in patient/mouse samples could not be tested with the available antibodies. DELE1/HRI loss and Sephin-1 treatment test protective stress-response context, not direct AFG3L2 cleavage of DELE1 or intrinsic transcription/kinase activity. |
| PMID:38012514 | Local full review, integrating earlier experiments across species. Protein quality control and substrate-processing context corroborate the established core; imprecise OMA1 ATPase/AAA terminology, MRPL32 degradation wording and calcium-direction summaries are not adopted as new mechanistic evidence. |
| PMID:38390227 | Local full diagnostic narrative review, although the immutable metadata labels it primary research. Its SCA28/SPAX5 inheritance, ocular findings and variant distribution are compiled clinical context, not an original protein-function assay. |
The original source routes are PubMed 36301938, PubMed 37804316, PubMed 38012514 and PubMed 38390227. The existing 77 source assertions, all action decisions and the integrated catalytic core remain unchanged. No new annotation is proposed from treatment rescue, disease association or downstream stress readouts.
The retained provider reports also cite Chandragiri and colleagues, AFG3L2-mediated proteolysis restricts mitochondrial biogenesis and gene expression in hypoxia, bioRxiv DOI 10.1101/2024.09.27.615438, posted 2024-09-30. The primary title/abstract and author publication listing identify a preprint; no exact PMID or later journal replacement was established. The abstract reports hypoxia-dependent proteolytic remodeling and a HIF1alpha–mTORC1 connection. Public full-HTML/PDF retrieval was unsuccessful; the correctly identified ResearchGate record also offers no full file. Its full Methods, substrate-specific controls and results were not inspected. It remains preliminary abstract-level research context, not an independently verified basis for new source-specific functions or an invented PMID/cache. The protected provider text remains unchanged.
The accepted manuscript for PMID:30252181 was reopened. It explicitly reserves all rights on PDF page 1 and repeated footers; the institutional record does not provide a contrary reuse license. Public accessibility therefore does not permit committing the full text. The existing short supporting_text_fulltext excerpt was reverified on manuscript page 16 (Results 3.3/Figure 3C). All three copies are preserved, with the original primary PMID; no self-citation to these notes is used as biological support. The local publication remains abstract-only. Human AFG3L2-V5 in m-AAA-deficient yeast, the protease-inactive E575Q control and the engineered Yta10–human paraplegin construct remain explicitly distinguished from native human organelle experiments.
Recursive scanning covers the authored review, these notes, historical PN notes and both immutable Falcon artifacts, including PMID/PMC links and normalized DOI URLs. All 25 identified required PMIDs and nine Reactome records are present. The one DOI-only preprint is separately identified and access-limited above, not silently counted as a recovered full paper. Four new reference assessments are added; prior reference identities and all published history remain intact. The old explicit PMID hyperlink is normalized to the renderer's bare PMID convention to remove its doubled URL display. YAML DRAFT is retained for any remaining validation advisories rather than implying unresolved normal-publication downloads.