ADCK5 is a 580-residue human member of the ancient UbiB (aarF-domain) family of atypical protein-kinase-like proteins, one of five human paralogs alongside ADCK1, ADCK2 and the two characterised coenzyme Q factors COQ8A (ADCK3) and COQ8B (ADCK4). In eukaryotes UbiB proteins are found exclusively in mitochondria and plastids, and ADCK5 is detected in high-confidence human mitochondrial proteome surveys and co-purifies predominantly with mitochondrial proteins; a hydrophobic segment at residues 50-67 is predicted to anchor it in a membrane. Its protein-kinase-like domain retains the complete catalytic and nucleotide-positioning core of the fold - the beta3 lysine, the alphaC glutamate, the catalytic-loop aspartate and asparagine, and the DFG aspartate - so it is not a degenerate pseudokinase. At the same time it retains both UbiB-specific features that distinguish this family from conventional kinases: the invariant KxGQ motif, whose N-terminal domain occludes the pocket that conventional kinases use to bind peptide substrates, and an alanine-rich loop in place of the canonical glycine-rich nucleotide loop, which in the characterised relatives produces an unusual preference for ADP over ATP. No catalytic activity, substrate or physiological process has been established for ADCK5 by direct assay. In the two relatives that have been assayed the fold does not support generic protein phosphorylation in trans, yet COQ8B does phosphorylate one specific partner protein, so a restricted, substrate-selective activity for ADCK5 remains an open possibility rather than an excluded one. Reported cellular roles - promotion of lung cancer cell invasion and migration via a SOX9/PTTG1 axis, and induction of airway epithelial senescence - rest on overexpression and knockdown studies without accompanying enzymology.
| GO Term | Evidence | Action | Reason |
|---|---|---|---|
| GO:0016020 membrane | IEA GO_REF:0000044 | ACCEPT | Summary: Correct but weakly grounded, and less informative than the available evidence. This row is the automatic mapping of UniProt SubCell SL-0162 (Membrane) to GO:0016020, and that SubCell assignment traces to a SUBCELLULAR LOCATION line reading "Membrane {ECO:0000305}; Single-pass membrane protein {ECO:0000305}" - a curator inference drawn from a purely predicted transmembrane helix at residues 50-67 (ECO:0000255). There is no experimental membrane measurement for ADCK5. The statement is nonetheless very likely true: UniProt records COQ8A as Mitochondrion membrane, single-pass, and ADCK1 as Mitochondrion, and ADCK5 is independently placed in the mitochondrion by the HTP row below. Kept as ACCEPT because a general parent term that is true should not be removed, and because refining it to GO:0031966 mitochondrial membrane would mean joining an experimental mitochondrion call to a predicted membrane call - a composite inference this review declines to make on ADCK5's behalf. The substantive issue is upstream and not fixable in GO: ADCK1 and ADCK2 receive SL-0173 (Mitochondrion) from the same UniProt pipeline while ADCK5 receives only SL-0162. That difference is not arbitrary - ADCK1 and ADCK2 each have an experimental mitochondrial localisation of their own (ECO:0000269|PubMed:33988507) that ADCK5 lacks - but the reason ADCK5 lacks it is that it was never tested: that study states its kinome library did not contain ADCK5. Raised as a UniProt correction request in suggested_questions, on ADCK5's own evidence rather than on parity with its paralogs. Reason: A true but prediction-derived parent term. Retained rather than removed because ADCK5 is almost certainly membrane-associated, and rather than refined because its submitochondrial placement has never been measured. Supporting Evidence: PMID:25498144 they contain predicted single-pass transmembrane (TM) domains that could limit solubility PMID:27499294 endogenous COQ8A is partially buried in the inner mitochondrial membrane with its C-terminus facing the matrix |
| GO:0005515 protein binding | IPI PMID:25416956 A proteome-scale map of the human interactome network. | MARK AS OVER ANNOTATED | Summary: Bare "protein binding" supported by a single yeast two-hybrid partner, NOTCH2NLA (Q7Z3S9), from the Rolland/CCSB proteome-scale interactome screen. Expanding the IntAct records shows this one screen is logged three separate times under three sub-method labels - two hybrid array, two hybrid prey pooling approach, and validated two hybrid - which is what inflates UniProt's NbExp to 4. It is one experiment, not three, and across all 54 IntAct records for Q3MIX3 there is no orthogonal assay for this pair (MI score 0.67 throughout). The compartments also sit awkwardly together: NOTCH2NLA is a reviewed 236-residue human-specific protein annotated Secreted and Cytoplasm that acts on neural progenitor proliferation, whereas ADCK5 is a mitochondrial UbiB-family protein whose only large interaction dataset returns 17 of 25 partners annotated to the mitochondrion. That compartment argument is a supporting consideration rather than a decisive one, and it is stated here as an assumption: it holds if ADCK5's kinase-like domain faces the matrix, as COQ8A's C-terminus is measured to do, but ADCK5's own sidedness has never been determined and an outer-membrane anchor presenting the domain to the cytosol is not excluded - which is the same uncertainty that stops this review proposing GO:0031966. The verdict does not depend on it: the method-replication argument stands alone. Marked as over-annotated rather than removed: this is an experimental IPI whose primary data this review has not seen, and the term is not positively contradicted, merely unreplicated by any independent method and uninformative as written. Not modified to a more specific molecular-function term because there is no characterised activity to modify it to. The partner-identity check ran clean - Q7Z3S9 is the reviewed, canonical, full-length Swiss-Prot entry, not a TrEMBL or partial-ORFeome substitute. Reason: One yeast two-hybrid screen triple-logged under three sub-method labels, with no orthogonal assay for this pair anywhere in IntAct, and a bare term that names no activity. Supporting Evidence: PMID:27499294 endogenous COQ8A is partially buried in the inner mitochondrial membrane with its C-terminus facing the matrix file:human/ADCK5/ADCK5-bioinformatics/RESULTS.md MI score 0.67 on every row; no orthogonal assay anywhere in IntAct. |
| GO:0005515 protein binding | IPI PMID:31515488 Extensive disruption of protein interactions by genetic vari... | MARK AS OVER ANNOTATED | Summary: Same partner as the row above, NOTCH2NLA (Q7Z3S9), from a second two-hybrid study that screens the effect of genetic variants on interactions. Judged separately per partner and per reference as required, and it reaches the same verdict for a reason specific to it: this is not independent corroboration of the Rolland hit. It is a two hybrid array experiment from the same CCSB resource lineage, i.e. method replication within one assay class rather than orthogonal validation. It inherits the same compartment consideration as the row above - a secreted and cytoplasmic human-specific neural regulator paired with a mitochondrial UbiB protein - and it inherits that consideration's limits too: ADCK5's membrane sidedness has never been measured, so this supports the verdict without being required by it. Had this been a different method, the two rows together would have justified MODIFY to an informative molecular-function term; as it stands there is no characterised binding activity to name. Over-annotated rather than removed, on the same grounds as the previous row. Reason: A second two-hybrid screen from the same CCSB resource lineage, so method replication within one assay class rather than the orthogonal validation this partner would need. Supporting Evidence: file:human/ADCK5/ADCK5-bioinformatics/RESULTS.md MI score 0.67 on every row; no orthogonal assay anywhere in IntAct. |
| GO:0005739 mitochondrion | HTP PMID:34800366 Quantitative high-confidence human mitochondrial proteome an... | ACCEPT | Summary: Well supported, and the single most informative annotation ADCK5 has. The reference is MitoCoP, a quantitative high-confidence human mitochondrial proteome, and the identical row is carried by ADCK1 and ADCK2. It is corroborated from three independent directions: the UbiB family is mitochondrial throughout eukaryotes; ADCK5's largest interaction dataset (PMID:27499296, the mitochondrial protein interaction map) returns 25 distinct partners of which 17 are UniProt-annotated to the mitochondrion; and the protein carries an N-terminal basic, hydroxylated-residue-rich segment of the kind that directs mitochondrial import. The complex-to-subunit projection failure mode was checked and does not apply here - PMID:34800366 is a proteome-wide localisation census that assigns no functional or phenotype term, so there is no phenotype able to spread across a complex, unlike the ComplexPortal-derived case documented on INO80 subunits. Reason: High-confidence mitochondrial proteome membership, corroborated independently by the family's exclusively mitochondrial distribution in eukaryotes and by an interactome dominated by mitochondrial proteins. Supporting Evidence: PMID:25498144 In eukaryotes, UbiB homologs are found exclusively in mitochondria |
Loading supporting content…
Download this section (compressed HTML)Q: UniProt correction request (1 of 2). Q3MIX3 carries EC=2.7.11.-, the Serine/threonine-protein kinase keyword, and consequently the cross-reference "GO:0004674; F:protein serine/threonine kinase activity; IEA:UniProtKB-KW", while the same entry states that the function of this protein is not yet clear and that it is not known whether it has protein kinase activity at all. UniProt has already applied exactly this correction to the two family members whose activity was measured: COQ8A and COQ8B were moved to the generic EC 2.7.-.- and lost the Ser/Thr-kinase keyword after Stefely et al. showed they lack canonical protein kinase activity in trans. ADCK5 and ADCK2 were not. Because GOA no longer imports keyword-derived annotations, today's GO record is unaffected - but any consumer reading UniProt directly still sees an activity assertion the entry's own FUNCTION line disclaims. Should ADCK5 be downgraded to EC 2.7.-.- with the Ser/Thr-kinase keyword removed?
Suggested experts: UniProt/Swiss-Prot curation
Q: UniProt correction request (2 of 2). Q3MIX3's SUBCELLULAR LOCATION is only Membrane, single-pass, inferred from a predicted TM helix, so its SubCell mapping is SL-0162 and it gains only GO:0016020 - no mitochondrial term. ADCK1 and ADCK2 both carry SL-0173 (Mitochondrion), each on an experimental annotation ECO:0000269|PubMed:33988507 from the subcellular kinome atlas. ADCK5 has no such annotation, and the reason is worth stating precisely: that study reports that ADCK5 was absent from its kinome library, so this is untested rather than tested-and-negative. Meanwhile ADCK5 has mitochondrial evidence of its own - it is in the MitoCoP high-confidence mitochondrial proteome (PMID:34800366), 17 of the 25 partners in its mitochondrial-interactome dataset are themselves mitochondrial, and every characterised eukaryotic UbiB protein is mitochondrial. Should Mitochondrion be added to ADCK5's SUBCELLULAR LOCATION on that basis? Posed on ADCK5's own evidence, not on parity with paralogs whose annotation rests on an assay ADCK5 never received.
Suggested experts: UniProt/Swiss-Prot curation
Q: PAINT/PAN-GO coverage. ADCK5 is the only one of the five human UbiB genes with no evolutionary-model annotation at all; UniProt records that Q3MIX3 has 0 GO annotations based on evolutionary models. PTHR43173, the family containing both ADCK1 (SF19) and ADCK5 (SF28), has exactly one annotated node, PTN005148758, seeded by a single yeast protein (MCP2, SGD:S000004243, itself SF19) and carrying GO:0005743, GO:0007005 and GO:0055088. ADCK1 inherits all three; ADCK5 inherits none, because the node lies inside the SF19 clade. This is a reach/coverage gap rather than a mis-propagation, and it is deliberately raised as a question and not acted on: a single yeast seed is thin support for extending three terms across the whole family, and ADCK5 already has direct HTP evidence for the mitochondrion. Should a family-level PTHR43173 node carry the mitochondrial localisation that is true of every characterised eukaryotic UbiB protein?
Suggested experts: GO Consortium PAINT curation
Q: Should GO be able to record the UbiB/ADCK atypical-kinase architecture at all? There is currently no way to state "protein-kinase-like fold with an occluded substrate pocket and an ADP-selective nucleotide loop, catalytic activity undetermined". For ADCK5 the honest GO position is simply the absence of a molecular-function term, which is what GOA now has - but the same gap pushed COQ8A's annotation into the generic GO:0016301 kinase activity for what is really ATP hydrolysis. The sharpest evidence of the cost is on COQ8B, not on ADCK5: GOA carries both NOT|enables GO:0004672 (IDA, PMID:27499294) and enables GO:0004672 (IDA, PMID:38425362) for that gene, so read as terms alone it asserts and negates protein kinase activity at once. That is an expressivity limit, not a curation error - COQ8B's reviewer reconciled the two explicitly in free text, recording that the negation refers to general in-trans behaviour and does not contradict the specific COQ3 phosphorylation. The reconciliation is correct and simply cannot be said in terms: GO offers no way to express "no activity in trans toward general substrates, but yes toward this one specific substrate". Raised as an ontology-coverage observation, not a request to create a term for this gene, and not an action on COQ8B, which is out of scope here.
Suggested experts: GO ontology development
Experiment: Differential scanning fluorimetry (thermal shift) on purified ADCK5 lacking the predicted transmembrane segment, against a panel of ADP, ATP and non-hydrolysable analogues, exactly as was done for COQ8A. This tests the single best-founded prediction about the protein and would decide between GO:0043531 ADP binding and GO:0005524 ATP binding. This review proposes neither term, because neither has been measured for ADCK5.
Hypothesis: ADCK5 binds adenine nucleotides with the ADP-over-ATP preference that the retained A-rich loop alanine A209 confers in COQ8A and yeast Coq8p.
Type: biochemical assay
Experiment: ATPase assay on purified ADCK5 in the presence and absence of cardiolipin and CoQ biosynthetic intermediates. A positive result would distinguish a conserved family ATPase from a genuinely divergent activity, and would support GO:0016887.
Hypothesis: ADCK5 has ATP hydrolysis activity modulated by lipids and CoQ intermediates, as COQ8A does.
Type: enzyme assay
Experiment: The COQ8B precedent shows the productive experiment is not a generic peptide-substrate panel - which is what failed for COQ8A - but a search for a specific partner. Run an ATP-dependent phosphorylation screen with intact-protein mass spectrometry against ADCK5's own co-purifying mitochondrial partners (CHCHD2, CHCHD3, IMMT, PMPCB and the other mitochondrial hits of PMID:27499296), rather than a generic kinase assay.
Hypothesis: Like COQ8B, ADCK5 phosphorylates a specific mitochondrial partner protein rather than generic substrates.
Type: biochemical assay
Experiment: Mutate the A-rich loop alanine A209 to glycine and test whether nucleotide selectivity flips toward ATP and autophosphorylation is enabled, as A339G does in COQ8A and A197G in yeast Coq8p. A positive result would establish that the occluded-pocket mechanism operates in ADCK5, which is currently inferred from sequence alone.
Hypothesis: The KxGQ/A-rich autoinhibitory architecture is functional in ADCK5, not merely present in sequence.
Type: mutagenesis
Experiment: Carbonate extraction and protease-protection assays on isolated mitochondria, with submitochondrial fractionation. This is the measurement that would justify GO:0031966 mitochondrial membrane or GO:0005743 mitochondrial inner membrane in place of the current generic GO:0016020; its absence is why this review proposes neither.
Hypothesis: ADCK5 is anchored in the mitochondrial inner membrane via the predicted 50-67 helix, with the kinase-like domain facing the matrix, as COQ8A is.
Type: cell fractionation
Experiment: PMID:32277958 infers the Ser181 site from mutagenesis of the substrate plus a motif-match argument, with no in vitro kinase assay reported. Perform a recombinant ADCK5 plus recombinant SOX9 kinase assay with a catalytically dead control mutating the DFG aspartate D382, to distinguish direct phosphorylation from an indirect consequence of overexpression.
Hypothesis: The reported ADCK5-SOX9-PTTG1 pathway involves direct phosphotransfer by ADCK5 to SOX9 Ser181.
Type: biochemical assay
What is not known — curated, literature-grounded statements of the open unknowns (the inverse of core functions).
Gap: ADCK5 has never been assayed. Its catalytic activity, its substrate and its physiological process are all undetermined: whether it transfers a phosphate to a protein, to a small molecule, or hydrolyses ATP without transferring anything at all, and whether it prefers ADP or ATP, are open questions with no measurement on either side. UniProt states the gap in its own FUNCTION line, and the paper that resolved the equivalent question for COQ8A explicitly set ADCK5 aside.
OPEN BIOLOGY WHOLLY_DARK
What is known: Established: mitochondrial localisation (MitoCoP high-confidence proteome, HTP), an interactome dominated by mitochondrial proteins, and a UbiB-family protein-kinase-like fold that retains the complete catalytic and nucleotide-positioning core (beta3 K228, alphaC E281, catalytic-loop D360 and N365, DFG D382) behind both UbiB-specific features that occlude the conventional substrate route (KxGQ K147, A-rich loop A209). Not established: any measured activity, any nucleotide preference, any substrate, any physiological process, and the submitochondrial compartment and membrane topology. Note the fold argument constrains but does not close the question in either direction - COQ8A was shown to lack canonical protein kinase activity in trans, yet COQ8B phosphorylates COQ3, so an activity restricted to one specific partner remains live.
Significance: ADCK5 is Pharos Tdark and is the only one of the five human UbiB genes with no evolutionary-model (IBA) annotation at all, so it inherits nothing from its family and has contributed nothing to it. After this review its entire GO record is one prediction-derived membrane term, one high-throughput mitochondrion term and two bare protein binding rows from a single two-hybrid pairing - the deep-unknome signature. The two characterised relatives both turned out to do something the family name did not predict, which is a reason to expect the same here rather than to assume ADCK5 is a redundant paralog.
Provenance (the field's own admissions):
Gap: GO cannot express what is actually known about this protein's molecular architecture: a protein-kinase-like fold whose conventional peptide-substrate pocket is occluded by its own N-terminal KxGQ domain and whose glycine-rich nucleotide loop is replaced by an alanine-rich, ADP-selective one, with the catalytic core intact but no activity determined. The only faithful GO position available is silence, so an architecture that is conserved from bacteria to humans and is the defining feature of an entire kinase-like family is unrepresentable.
OPEN ONTOLOGY MF_DARK
What is known: Established: the UbiB-specific features are defined, quantified across the family, and structurally interpreted, and ADCK5 retains them. Missing: any term between GO:0016301 kinase activity, which asserts an activity ADCK5 has not been shown to have, and no term at all. The cost is visible on the characterised paralogs too - COQ8A's demonstrated ATP hydrolysis is recorded as generic GO:0016301 kinase activity alongside a NOT annotation to GO:0004672, which is an accurate but awkward way to describe an enzyme, and COQ8B ends up carrying enables GO:0004672 and NOT|enables GO:0004672 at the same time because the ontology cannot distinguish a generic activity from a single-substrate one. In COQ8B's case the curator reconciled the two correctly in prose; the point is that no arrangement of GO terms could have carried that reconciliation.
Significance: This is a family-level gap rather than an ADCK5-specific one; it affects all five human ADCK genes and their orthologs throughout eukaryotes and bacteria. No new term is proposed here, because for ADCK5 specifically there is no measured activity for such a term to record - the gap is raised so that it is visible when a member is eventually characterised.
Provenance (the field's own admissions):
Loading supporting content…
Download this section (compressed HTML)Loading supporting content…
Download this section (compressed HTML)Loading supporting content…
Download this section (compressed HTML)Loading supporting content…
Download this section (compressed HTML)