SIRT5: paired evidence notes

Human reference: https://www.uniprot.org/uniprotkb/Q9NXA8/entry

Primary source: PMID:22076378 Sirt5 is a NAD-dependent protein lysine demalonylase and desuccinylase..

We found that Sirt5
is an efficient protein lysine desuccinylase and demalonylase in vitro. The
preference for succinyl and malonyl groups was explained by the presence of an
arginine residue (Arg(105)) and tyrosine residue (Tyr(102)) in the acyl pocket
of Sirt5.

The horse sequence comparison is in SIRT5-bioinformatics/RESULTS.md. Transfer is assessed for this exact accession; human biochemical evidence is not equine experimental validation.

Horse-specific research access

A horse-specific Falcon/Edison investigation was requested because PMID:36361948 studies equine chondrocytes and reports SIRT5-expression changes. Edison rejected task creation with HTTP429; the configured Perplexity fallback returned HTTP401 insufficient quota. No horse provider report was generated. Expression-level evidence does not resolve the divergent F6S899 protein model or validate catalytic activity.

2026-10-20 — Full re-review, independent of the ProtNLM2 predictions

Scope: re-examined the single GOA row (GO:0070403 NAD+ binding, IEA, InterPro IPR003000), the description and
core functions. ProtNLM2 files were not opened.

Identity of F6S899: a defective gene model of horse SIRT5

New analysis: SIRT5-bioinformatics/compare_gene_models_rereview.py -> RESULTS-rereview.md /
results-rereview.json (live UniProt, Ensembl REST and QuickGO; the original RESULTS.md is untouched).

• Horse gene ENSECAG00000026898 has several protein-coding transcripts. F6S899 = ENSECAT00000028843 (282 aa).
  Full-length alternatives: A0A9L0T9B1 = ENSECAT00000146678 (310 aa, 88.1% identical to human Q9NXA8 over its whole
  length) and A0A9L0T6A9 = ENSECAT00000079525 (318 aa, the Ensembl canonical transcript, alternative C-terminal exon).
• F6S899 matches human SIRT5 window by window through human residue ~200 (identical to the other horse models), then
  collapses (10/30 and 7/40 identical in human windows 201-250 and 251-300).
• Cause: the F6S899 transcript omits the internal exon 13742662-13742785 (124 nt, 124 mod 3 = 1) present in both
  full-length transcripts, so everything downstream is translated in the wrong frame. Its last two "exons" are
  separated by a 3-nt gap (13757851-13757853), another sign of a computational model rather than a real splice form.
• Consequence for mapped human features: retained His158 (catalytic base), NAD loop 58-77, Tyr102/Arg105 acyl pocket,
  QNID 140-143, Cys166/Cys169; lost Cys207/Cys212 (zinc; W/W), GTS 249-251, NTE 275-277, Cys293. The full-length horse
  models keep all of these (A0A9L0T9B1 has NME for NTE, as many mammals).
• UniProt CAUTION on F6S899 ("Lacks conserved residue(s) required for the propagation of feature annotation",
  PROSITE PRU00236) and the failure of HAMAP MF_03160 to match F6S899 (it matches A0A9L0T9B1) are consistent with this.

Evolutionary grounding

• PANTHER: F6S899 and A0A9L0T9B1 are both PTHR11085:SF10 (SIRT5 subfamily). The family review
  interpro/panther/PTHR11085/PTHR11085-review.yaml lists F6S899 and human Q9NXA8 as SF10 representatives; no
  disagreement about subfamily placement. (The family review says "no catalytic-residue loss is inferred from a short or
  incomplete input sequence"; this re-review identifies the cause of the incompleteness as a gene-model defect.)
• PAINT: node PTN000119399 (IBD seeds MGI:1915596 Sirt5 and UniProtKB:Q9NXA8) carries GO:0036054, GO:0036055,
  GO:0061697, GO:0005739, GO:0005829; PTN000872307 carries GO:0005759. QuickGO shows these IBAs on A0A9L0T9B1, not on
  F6S899. That is PAINT's one-representative-per-gene propagation, not a functional judgment against the gene; but it
  means the curated record for this gene's functions is A0A9L0T9B1.

Decisions

• GO:0070403 NAD+ binding: UNDECIDED -> KEEP_AS_NON_CORE. True of the horse SIRT5 gene product (NAD+ is the
  co-substrate), domain-signature based; not the core activity. Whether the truncated F6S899 translation binds NAD is
  not established, stated in the reason.
• core_functions: added desuccinylase (core), demalonylase and deglutarylase activities in mitochondrial matrix, scoped
  to the gene and grounded in the SF10/PTN000119399 placement plus the mammalian primary literature
  PMID:22076378
  PMID:24703693.
  No NEW existing_annotations rows were added to F6S899 itself, because its own translation lacks the second zinc
  pair; the right anchor for those terms is a full-length product (A0A9L0T9B1 already carries them by IBA).
• Description rewritten as a project-independent summary that also records the gene-model defect.

Relative to review next

Benchmark accession is an alternative/defective gene model; no reviewed (Swiss-Prot) horse SIRT5 exists, and no repo
review exists for the full-length horse product. Relative: A0A9L0T9B1 (same gene, full-length 310-aa product that
carries the PAINT IBAs) -> folder genes/HORSE/SIRT5__A0A9L0T9B1. Human SIRT5 (Q9NXA8) already has a review
(genes/human/SIRT5/SIRT5-ai-review.yaml, all 68 rows actioned, core functions present).

Modules: no module in modules/ includes SIRT5 (grep for SIRT5/Q9NXA8 found none; desuccinylase hits are bacterial
succinylarginine/lysine-biosynthesis modules, unrelated).