The initial entries below are preserved as a curation journal. The review follow-up at the end records the current disposition where it supersedes an earlier decision.
Human FITM2 is UniProtKB Q8N6M3, HGNC:16135, also called FIT2 or C20orf142. The genuine seed contains 35 annotation contexts and no natural alternative protein products. All incoming term IDs, evidence codes, references, qualifiers and supporting entities are preserved. The canonical sequence has 262 residues. Engineered catalytic or binding mutants, disease-associated truncations, the mouse ortholog and the HTATIP2 partner isoform are not recorded as alternative FITM2 products.
The review contains 26 ACCEPT, eight KEEP_AS_NON_CORE and one UNDECIDED decisions, two core functions and no NEW annotations. Support is reference-only, with no new publication quotations. The linked evidence record contains exact target ancestry, original IBD joins, interaction records, source hashes and access boundaries. The review cites both this guide and that record through file:human/FITM2/... references.
PMID:32915949(https://pubmed.ncbi.nlm.nih.gov/32915949/) provides the decisive biochemical evidence. Its actual Results and construct/purification Methods describe recombinant human FIT2, wild type and catalytic-histidine mutants, radiolabeled substrate assays and product mass spectrometry. FIT2 cleaves fatty-acyl-CoA into acyl-4′-phosphopantetheine and adenosine 3′,5′-bisphosphate. Another microsomal activity produces the subsequent acyl-pantetheine product; that second reaction is not assigned to FITM2. The measured substrate preference does not establish equal activity on every CoA derivative. The ER-luminal orientation of the catalytic region is a topology-based interpretation, not a direct localization measurement from the purified assay.
The official GO:0010945 definition explicitly includes acyl-CoA substrates. Its short label, coenzyme A diphosphatase activity, therefore describes the measured reaction. A generic lipid-phosphatase-family prediction is not substituted for it. Human cellular complementation connects active FIT2 to ER morphology and normal lipid-droplet formation. Long-term changes in triglyceride synthesis and associated expression were not fully rescued and are separated from the immediate catalytic mechanism. Figure captions and text were read; figure pixels and separate supplements were not independently audited.
PMID:22106267(https://pubmed.ncbi.nlm.nih.gov/22106267/) supports direct detergent-solubilized FIT2 binding to triolein, saturation/competition behavior and binding-mutant comparisons. Accessible indexed Results also describe DAG binding. The official mouse donor P59266 is Fitm2/Fit2 and links this study and N80A perturbation to binding/droplet phenotypes. Complete original Methods and the species of every construct were not recovered; the incoming ISS/IEA contexts are retained as transfers. No purified human binding assay is claimed solely because later literature describes the earlier work that way.
The second core function summarizes triglyceride binding in lipid storage, with associated DAG-binding capacity. Its mechanistic relationship to acyl-CoA diphosphatase activity remains a question, not an established reaction sequence. DAG binding is not DAG hydrolysis.
Reactome R-HSA-8857686 describes FITM2 binding TAGs and partitioning them into droplets. The official GO:0170007 definition instead requires bringing an ER membrane and a lipid droplet together to establish a contact site. The accessible binding/partitioning experiments do not settle physical membrane bridging. The TAS tether annotation is UNDECIDED, with incomplete full-text access disclosed; absence of a visible assay is not used to REMOVE it. The reaction's ER location is supported independently. Its apparent repeated-paralog-name typo is not reproduced as biological fact or used to dismiss the reaction.
The genuine normal fetch recovered the HTML full text of PMID:18160536(https://pubmed.ncbi.nlm.nih.gov/18160536/). Mouse membrane fractionation, mouse constructs expressed in HEK293 cells, tracer partitioning and murine 3T3-L1 shRNA Results were read with the available Methods. Human host cells do not convert the mouse constructs into human protein experiments. FIT2 partitioned newly synthesized triglyceride into droplets without increasing synthesis in the overexpression system. In differentiating knockdown adipocytes, triglyceride biosynthetic flux decreased. The existing broad triglyceride-metabolic involvement is retained as non-core, without calling FITM2 a triglyceride synthase or lipase.
The same original paper reports reduced PPARγ, aP2 and adiponectin in the strongest shRNA1 condition, while shRNA2/3 retained broadly similar marker expression despite reduced droplets. Neither uniformly preserved nor universally blocked differentiation describes those experiments. The two existing differentiation transfers are retained as non-core and are not promoted to direct regulation of adipogenic commitment.
PMID:24519944(https://pubmed.ncbi.nlm.nih.gov/24519944/) used a later, aP2-driven mouse deletion. The conditional adipocytes retained differentiation markers while producing fewer, larger droplets; this system cannot exclude an earlier developmental role. Its normal cache remains abstract-only; bounded independent primary consultation read the official PMC Methods, Results and captions. It is corroborating context rather than a reason to erase the strongest-shRNA observation.
The complete official PANTHER PTHR23129 topology has 157 nodes. The exact human leaf is PTN002508545, named HUMAN|HGNC=16135|UniProtKB=Q8N6M3; its ungapped MSA sequence exactly matches canonical UniProt. All four incoming IBA origins match positive original IBD assertions on that target path: ER membrane, lipid homeostasis and lipid-droplet formation at PTN000565881, and CoA diphosphatase at PTN002751040. Human FITM2 among experimental descendant seeds is legitimate and is not treated as circular. The family fat-cell-differentiation IBD at PTN002751037 is off the human path and is not imported. That does not refute the separate mouse-based ISS/IEA contexts.
The six-row normal PAINT slice was regenerated with the repository parser/serializer from the actual complete family-node set and the genuinely fetched original IBD GAF already shared by the campaign. No bulk leaf GAF was downloaded. Release-19 API topology and the original PAINT data have separate provenance; this check does not assert whole-release equivalence. Source URLs, hashes, node joins and replay method are recorded in the evidence YAML.
All four generic interaction contexts were matched to official IntAct/IMEx records. PMID:28514442(https://pubmed.ncbi.nlm.nih.gov/28514442/) and PMID:33961781(https://pubmed.ncbi.nlm.nih.gov/33961781/) support FITM2-TSPAN18 association by anti-tag coimmunoprecipitation in human HEK293T cells. PMID:32296183(https://pubmed.ncbi.nlm.nih.gov/32296183/) supports UNC50 and HTATIP2-3 associations in yeast two-hybrid systems. The three method records for each HuRI pair are not counted as three biological replications. Q9BUP3-3 is the partner product; its reconstructed 276-residue sequence matches the official FASTA and its recorded 35–276 region matches canonical HTATIP2. No isoform-exclusive interaction, adaptor activity or partner function is assigned to FITM2.
The explicit campaign policy retains these four supported generic-binding rows as non-core when no informative subtype is established. This is an exception to the default preference against protein binding; the validator is not changed or suppressed.
PMID:21834987(https://pubmed.ncbi.nlm.nih.gov/21834987/) identifies C20orf142/FITM2 among morphology-screen hits. The phenotype does not establish a defining cytoskeletal mechanism. Its correction, PMID:39390519, replaces duplicated siARC/siZRANB Figure 6 images and reports unchanged conclusions. The notice is neither a new FITM2 assay nor a retraction; no independent image-forensics claim is made.
PMID:28067622(https://pubmed.ncbi.nlm.nih.gov/28067622/) and PMID:30214770(https://pubmed.ncbi.nlm.nih.gov/30214770/) support the human recessive deafness-dystonia association. Disease phenotypes and fly knockdown corroboration are separated from human biochemical evidence. PMID:30288795 has metadata but no substantive accessible abstract/body in the cache, so variant-specific effects remain unverified. PMID:11780052 and PMID:15489334 provide sequence/cDNA provenance, not catalytic evidence.
The review, annotation-reviewer and core-function-synthesizer workflows were applied with a separate read-only annotation consultation. Prior genuine Falcon authorization and fallback quota failures remain unresolved; no provider-branded research file was fabricated and no repeated credential-failing request was made. Manual research and its access limits are documented here.
The candidate began at main 2cbe66960f237b9ed3c937642e5a6a918662b200 after a complete, stable 257-open-PR ownership check. Nineteen normal source paths were qualified: eight exact main sources and eleven absent, unowned paths. An inherited sparse view initially omitted a qualified main publication, so title acquisition refetched it. The fresh response was preserved privately and the exact original main bytes restored immediately; all nineteen qualified paths were then included in the sparse view and all eight main sources verified unchanged. The operating rule for reuse is to hydrate qualified main sources before fetching. Prior task branches, 41 inherited scratch files and the shared runtime are preserved; no environment or worktree was created and no old temporary material was cleaned.
Two independent readers found no material scientific issue in the 35 decisions and two core functions. Targeted schema, ontology-ID/label, reference-title, strict GOA, history and status checks passed, and a private HTML preview rendered successfully. The gene validator reports four advisory warnings for the deliberately retained generic-binding rows, so the computed status is DRAFT. No rule was weakened. History schema/backfill tests passed after hydrating their three small, unchanged main fixtures omitted by the sparse view; the initial missing-fixture failures were retained in the private audit trail. No general application test expansion or source-cache change was needed.
Re-reading the Results together with the Discussion of PMID:18160536(https://pmc.ncbi.nlm.nih.gov/articles/PMC2224239/) supports marking both incoming GO:0045444 fat-cell-differentiation transfers as MARK_AS_OVER_ANNOTATED. The authors interpret the principal phenotype as impaired triglyceride-droplet accumulation rather than differentiation. The strongest shRNA1 condition did reduce PPARγ, aP2 and adiponectin; that observation is preserved alongside the broadly retained markers in shRNA2/3. Severe storage impairment could secondarily affect those markers, but this explanation was not established experimentally. The later aP2-driven deletion in PMID:24519944 cannot establish or exclude an earlier role. The off-path family IBD is not independent evidence against the separate mouse transfers. The current review therefore has 26 ACCEPT, six KEEP_AS_NON_CORE, two MARK_AS_OVER_ANNOTATED and one UNDECIDED decisions, with all 35 incoming contexts preserved.
DAG-binding and TAG-competition details for PMID:22106267 were read in externally accessed indexed PMC3241795 Results, not recovered from its abstract-only normal cache. Complete original Methods and all construct provenance remain unavailable. PMID:32915949 describes the earlier purified protein as human; that later account does not substitute for auditing every original construct. The two core functions remain supported, and the unresolved physical ER–droplet tether remains UNDECIDED.
Explicit user campaign instructions retain supported generic binding as non-core when no informative subtype is established and require new support to remain reference-only without publication quotations. These instructions take precedence over conflicting local skill and CLAUDE.md defaults; they are not a legacy-content exception. The four association contexts and source-specific evidence remain documented, and default validators remain unchanged.