Run 2026-09-26; residue-159 check added and re-run 2026-10-10. Script: analyze_ak4p3.py (no dependencies; sequences fetched live from the
UniProt REST API, comparison and digest computed — nothing hardcoded). Raw output:
run-output.txt.
AK4P3 carries 19 GO annotations describing a working mitochondrial adenylate kinase — AMP
kinase activity, nucleoside triphosphate adenylate kinase activity, mitochondrial matrix, ATP/GTP
metabolic process. Every one is IEA or IBA, applied by automatic pipelines (chiefly HAMAP-Rule
MF_03170) on similarity to AK4.
But HGNC names the locus "adenylate kinase 4 pseudogene 3", locus_type: pseudogene. Those
cannot both be casually true.
| Test | Finding |
|---|---|
| ORF length | 223 aa — identical to AK4 (P27144) |
| Identity to AK4 | 221/223 (99.1%) |
| Differences | two, both conservative: M53V and V177A |
| Catalytic P-loop (Walker A) | GPPGSGK at position 12 — intact in both |
| Residue 159 (AK4's inactivating Gln) | Q in both (context LVQQEDD) |
| Tryptic peptides unique to AK4P3 (≥7 aa) | two: ASTEVGEVAK, DAAKPVIELYK |
The sequence gives no evidence of pseudogenisation at the protein level. The ORF is intact,
full-length, and near-identical to AK4, with the catalytic P-loop conserved. The automatic
pipelines are not misfiring on a decayed relic the way they do for, say, a frameshifted remnant.
But AK4P3 also keeps AK4's Gln159. Liu et al. (PMID:19130895) found AK4 inactive as an
adenylate kinase in vitro and traced this to a glutamine at 159 where other adenylate kinases have
a conserved arginine; Q159R restored activity with GTP. AK4P3 has the same glutamine, so its
sequence predicts AK4's enzymology, not that of an active adenylate kinase. AK4's own activity is
contested in the literature (another group measured AMP phosphorylation by recombinant AK4,
PMID:19766732), so the sequence settles which enzyme AK4P3 resembles, not whether that enzyme is
an AMP kinase.
The genuine uncertainty is one level up and outside GO's reach: is this locus transcribed and
translated? HGNC says pseudogene; Ensembl gives the same locus biotype: protein_coding while
keeping the pseudogene name; UniProt has it unreviewed (TrEMBL) at PE=1, evidence at protein
level, with a PeptideAtlas cross-reference.
The PE=1 is checkable, which is the useful part. Peptide-to-protein assignment between
99%-identical paralogs is exactly where proteomic evidence goes wrong — but here two fully tryptic
peptides are unique to AK4P3, and both substitutions are mass-distinguishable (M→V is −32 Da,
V→A is −28 Da). So the evidence can be attributed correctly. Whether it was depends on whether
those two peptides are the ones observed, which this analysis cannot determine.