AFF4-CDK15: AFF4 is prey in a systematic CMGC-kinase TAP interactome - a kinase adjacent to the complex CDK9 works in, and nobody has followed it up
also called AFF4-CDK15 · AFF4-ALS2CR7 · CMGC kinase interactome
A systematic survey of which proteins stick to each human CMGC-family kinase found AFF4, the scaffold of the elongation complex, attached to the poorly studied kinase CDK15. Nobody has looked at it since.
The same thing, for a scientist
**Verdict: ADMITTED AS A LEAD, on relevance rather than on strength.** IntAct holds one AFF4-CDK15 record: **TAP**, *association*, HEK293, AFF4 as prey, intact-miscore 0.35, from **PMID 23602568** [E1] - Varjosalo, Keskitalo, Van Drogen, Nurkkala, Vichalkovski, Aebersold and Gstaiger, *Cell Reports* 2013, a global proteomic analysis of complexes of the human **CMGC kinase group**, reporting **652 high-confidence kinase-protein interactions** and stating explicitly that its purpose is to give *a specific biochemical context for many poorly studied CMGC kinases* [E2]. This is a **focused family-scale survey, not a proteome-scale atlas**: the baits are one kinase family and the design is systematic. **The reason to keep it is where it sits.** The super elongation complex's business end is **CDK9**, a CMGC-adjacent kinase this graph already carries as `gene-cdk9` and `protein-ptefb-cdk9-ccnt1`. A second, orphan CDK recovered on the AFF4 scaffold is the kind of lead that would be obvious if anyone had looked, and CDK15 appears exactly once in this corpus before now - inside a list of unwritten partner symbols in the King's tail-ordering analysis, enumerated and never discussed [E1].
What the row supports and what it does not
Supports: in HEK293, a TAP purification through CDK15 recovers AFF4. TAP is a two-step purification and is more stringent than single-step anti-tag co-IP, which is the main reason this row is treated better than several others in the census.
Does not support: that AFF4 is a CDK15 substrate; that the contact occurs in a haematopoietic cell; that CDK15 is in the SEC; that CDK15 phosphorylates anything relevant. The paper's own framing is that it provides biochemical context for poorly studied kinases and identifies candidate substrates - candidate, from the abstract.
The honest size of the claim: one TAP row in embryonic kidney, from a survey whose value is breadth. CDK15 (Q96Q40) is among the least-characterised human CDKs, which cuts both ways: it is unstudied, so an interaction is more likely to be genuinely novel, and it is unstudied, so there is no independent literature to corroborate against.
Why the CDK9 adjacency is a reason to look and not a reason to believe
This corpus's elongation story is: KMT2A::AFF1 recruits the super elongation complex; AFF1/AFF4 scaffold it; P-TEFb (CDK9-cyclin T1) phosphorylates the Pol II CTD, NELF and DSIF and releases paused polymerase. CDK9 inhibitors are a live therapeutic direction and gene-cdk9 was written precisely because a database diff found it to be the best-evidenced partner with no entry of its own.
A second CDK on the same scaffold invites the obvious hypothesis and the obvious trap. The hypothesis is that CDK15 contributes a phosphorylation the field attributes to CDK9. The trap is that CDKs are a family with conserved folds and shared interaction surfaces, so recovering one CDK on a scaffold that binds another is close to the null expectation for a family-scale survey - and this project has been caught four times reporting a result that a base rate produces (rule 84).
The test that distinguishes them is cheap and is not an experiment: the same publication purified every CMGC kinase. How many of them recover AFF4? If most do, the row is a property of the scaffold's stickiness; if CDK15 is one of one or two, it is a lead. The data are in the paper's supplementary tables and were not opened here. That null is named and unrun, deliberately, rather than the row being scored without it.
| partner a | gene-aff4 |
|---|---|
| partner b | CDK15 |
| partner b uniprot | Q96Q40 |
| partner b length AA | 435 |
| partner a uniprot | Q9UHB7 |
| partner a length AA | 1163 |
| verdict | ADMITTED AS A LEAD |
| adjudicated by | q160, cycle 146, 2026-09-24 |
| druggability | CDKs are the most tractable kinase subfamily in oncology, but CDK15 has no selective chemical probe in this corpus and no assigned substrate. Any CDK9 inhibitor used in this disease is likely to hit it without anyone knowing. |
| n methods | 1 |
| n platforms | 1 |
| n publications | 1 |
| endogenous side | no - tagged CDK15 bait |
| intact miscore | 0.35 |
| cell state | unstimulated, asynchronous |
| epigenetics | None directly; the relevance is transcriptional, through the super elongation complex. |
| named unrun null | How many of the CMGC kinases purified in PMID 23602568 recover AFF4? If most do, this row is scaffold stickiness. The data are in that paper's supplementary tables and were not opened. |
| absence class | never-measured |
| absence class note | No experiment has tested CDK15 against AFF1, against a KMT2A fusion, or in any haematopoietic cell, and no substrate is assigned to it. |
interface residues a
[]
structures
[]
ligands
[]
records
[ "AFF4-CDK15, tap, association, HEK293, AFF4 prey, miscore 0.35, PMID 23602568" ]
cell systems
[ "HEK293 embryonic kidney" ]
localization
{
"compartment": "not recorded in this row",
"subnuclear": [],
"shown_for_fusion": "NO - nobody has looked in any fusion."
}measurements
[
{
"quantity": "count",
"subject": "entries in this corpus naming CDK15 before this beat",
"value": 1,
"units": "entries",
"n": 2931,
"universe": "word-boundary scan of all node files excluding q160, 2026-09-24",
"null": null,
"note": "CORRECTED before shipping: this entry first said zero. CDK15 appears once, inside a list of unwritten partner symbols in ana-the-interactome-tail-is-orderable-after-all-by-cost-to-complete-and-not-by-evidence-strength. It has been enumerated, never discussed."
}
]- If an unstudied CDK sits on the SEC scaffold, then every CDK9-inhibitor phenotype in this disease has an unexamined off-target explanation - and the inhibitors are being developed on the assumption that CDK9 is the relevant kinase on that complex.
- Admitting a row because the partner is in an interesting family is the base-rate error rule 84 records. This entry names the null rather than computing it, which is an improvement on not naming it and is not a substitute for running it.
- How many CMGC kinases in PMID 23602568 recover AFF4? This is the null for the row and it is one supplementary table away.
- Do the CDK9 inhibitors used in this disease inhibit CDK15? Kinome selectivity panels exist for most of them and CDK15 is usually in them; nobody has looked because CDK15 has no known function here.
- Is AFF1, the partner actually present in the index fusion, also recovered by any CMGC kinase? The survey covered the family, not the scaffolds, so the AFF1 answer is in the same tables.
- E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database
IntAct REST returns one AFF4-CDK15 (Q9UHB7-Q96Q40) record: detection method 'tap' (tandem affinity purification), interaction type 'association', host Homo sapiens HEK293 embryonic kidney cell, AFF4 as PREY, intact-miscore 0.35, from PMID 23602568.
IntAct molecular interaction database, REST findInteractions on Q9UHB7, retrieved 2026-09-24 by q160 (partner CDK15) · IntAct (EMBL-EBI) 2026What this evidence cannot show
One record. TAP is two-step and therefore more stringent than single-step co-IP, but it is still an over-expressed tagged bait in embryonic kidney. The miscore of 0.35 is the floor for a single-record association and does not reflect the method's stringency.
Evidence that argues against this (1)
- The same survey deposited many rows; without knowing how many CMGC kinases recover AFF4, a single row cannot be distinguished from a scaffold that is recovered by the whole family. That control exists in the source's supplementary tables and was not run. Varjosalo M et al. 2013
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
PMID 23602568 (Varjosalo et al., Cell Reports 2013) presents a global proteomic analysis of complexes of the human CMGC kinase group, identifying 652 high-confidence kinase-protein interactions, reporting subgroup-specific functional enrichment and modularity, a kinase-kinase subnetwork, candidate substrates for CMGC kinases, and a large set of interactions with proteins genetically linked to human diseases including cancer. Its stated purpose includes providing specific biochemical context for poorly studied CMGC kinases. AFF4 and the super elongation complex are not named in the abstract.
The protein interaction landscape of the human CMGC kinase group. · Varjosalo M, Keskitalo S, Van Drogen A, Nurkkala H, Vichalkovski A, Aebersold R, Gstaiger M · Cell Rep 2013What this evidence cannot show
Abstract read in full; the 652-interaction table was not opened, so neither the AFF4 row's score within the study nor the number of CMGC kinases recovering AFF4 is known here. 652 interactions across the CMGC group is a focused family survey rather than a proteome-scale atlas, which is why this beat's >=15-partner atlas heuristic does not flag it and why that non-flagging is, in this case, correct.
Evidence that argues against this (1)
- 'High-confidence' in a systematic AP-MS survey is a statistical threshold applied across thousands of candidate pairs, not a judgement about any individual pair. A row at the threshold in a 652-interaction set is not equivalent to a hypothesis-driven co-IP. Varjosalo M et al. 2013
AFF4 prey in a CDK15 tandem affinity purification, HEK293
a second, orphan CDK on the scaffold whose established kinase is CDK9 - and the null (how many CMGC kinases recover AFF4) is named and unrun
one of the 135 uncovered human prey-only partners