Oncofusion

start / graph / mechanism-bmi1-cdkn2a-cdk6-axis

Mechanismmoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q56-homeodomain

BMI1, CDKN2A and CDK6 are one circuit, not three genes - and in KMT2A-rearranged disease it has a loop nobody has measured

also called polycomb-INK4a-CDK6 axis · BMI1-p16INK4a-CDK4/6 · INK4a/ARF locus control

One gene, CDKN2A, makes the brake on cell division; BMI1's job is to hold that brake off and CDK6 is what the brake grips, so the three are a single loop rather than three separate targets. In KMT2A-fusion leukemia the brake is usually still physically present and merely held down, which is the situation where the loop matters most and where nobody has measured it.

The same thing, for a scientist

The circuit runs: **BMI1 (PCGF4/canonical PRC1) represses the *CDKN2A/CDKN2B* locus -> less p16INK4a -> less inhibition of CDK4 and CDK6 -> RB phosphorylation and G1 exit.** Three things make this more than a textbook diagram in KMT2A-rearranged leukaemia. **(1) The fusion has its own way in.** Hoxa9 regulates the same *p16Ink4a/p19Arf* locus as Bmi1 and substitutes for it: MLL-AF9 transforms *Bmi1*-null murine cells readily while non-Hox-activating fusions trigger senescence there [E1]. **(2) The circuit is a loop, not a chain.** CDK6 itself transcriptionally *induces* p16INK4a in a **kinase-independent** complex in murine BCR-ABL+ B-ALL, so CDK6 sits both downstream and upstream of the brake [E2]. **(3) The brake is usually intact.** *CDKN2A* is deleted in only about 8.5% of infant KMT2A-rearranged ALL against 28% of infant non-KMT2A ALL [E3], so in most patients the repression is epigenetic and in principle reversible. Live DepMap sets the constraint on any of it: CDKN2A knockout is fitness-favourable (+0.157), BMI1 is neutral (+0.0585), RB1 is favourable (+0.1801), and only CDK6 is a real dependency in leukaemia lines (-1.3767) [E4].

Go deeper
The circuit, and what is measured at each step in THIS disease
stepwhat is establishedmeasured in infant B-lineage KMT2A::AFF1?
BMI1 represses CDKN2A/Bestablished across the polycomb field; the effector of BMI1's self-renewal role 5no - not found this cycle in any KMT2A-rearranged system
HOXA9 also regulates p16Ink4a/p19Arf and substitutes for Bmi1murine MLL-AF9 1no - murine, myeloid
CDKN2A encodes p16INK4a, the inhibitor of CDK4 and CDK6established; UniProt P42771's own alternative name is cyclin-dependent kinase 4 inhibitor An/a
CDKN2A is usually genomically intact here4/47 deleted (8.5%) vs 5/18 (28%) in infant non-KMT2A ALL, P=0.058 3yes - this is the one human infant measurement in the circuit
CDK6 is required, CDK4 is nottwo knockdown studies plus a genome-wide screen (gene-cdk6)partly - one arm used SEM, a t(4;11) line from a five-year-old
CDK6 induces p16INK4a kinase-independentlymurine BCR-ABL+ B-ALL 2no

Five of six rows are unmeasured in the index disease. The circuit is asserted throughout this literature and assembled here from four systems, none of them infant B-lineage KMT2A::AFF1 in human cells. That is the honest status and it is why this is a mechanism entry rather than a target entry.

The loop, which is the part that changes what an intervention does

Drawn as a chain - BMI1 down, p16 up, CDK6 off - the circuit implies that a BMI1 agent and a CDK6 agent do the same job and are interchangeable.

Drawn as a loop it does not. CDK6 is part of a transcription complex that induces p16INK4a, and that function is independent of its kinase activity 2. The authors' own reading is that high CDK6 suppresses proliferation by upregulating p16INK4a as an internal safeguard, and exerts its full tumour-promoting function only when p16INK4a is absent - which is why they propose a selection pressure to raise CDK6 and silence p16INK4a together.

Two consequences follow and both are testable:

  • An ATP-competitive CDK4/6 inhibitor removes the kinase arm and leaves the p16-inducing scaffold, so it is not a CDK6 knockout and its effect on the brake is unpredictable from the genetics. This is the specific break in the knockout-to-drug inference recorded in gene-cdk6.
  • In a cell where CDKN2A is intact - about 91% of infant KMT2A-r patients 3 - the loop is closed. Inhibiting CDK6 catalytically could relieve the very p16 induction that limits proliferation, which is a mechanism for a CDK4/6 inhibitor to underperform its genetic prediction. Nobody has measured p16INK4a protein under a CDK4/6 inhibitor in a KMT2A-rearranged cell in anything read this cycle.
Why the deletion frequency is the most useful number in the circuit

Most of what is known about CDK4/6 inhibitors clinically comes from tumours where CDKN2A is lost, which is the canonical enrichment biomarker for the class. Infant KMT2A-rearranged ALL is the opposite case: the locus is intact in roughly nine patients in ten 3.

That cuts both ways, and the graph should not pick a side it cannot support:

  • Argument that the class should work anyway. RB1 is intact too, and RB1 function is the actual requirement for CDK4/6 inhibitor response. Live DepMap gives RB1 a positive gene effect (+0.1801 across 1,258 lines), consistent with RB1 being an intact brake these cells tolerate rather than a gene they need.
  • Argument that it should not. An intact p16INK4a means an endogenous CDK4/6 inhibitor is already present and the cells proliferate anyway - which says the pathway is not rate-limited where a drug would act, or that the repression is strong enough to overwhelm it. A cell that has arranged to ignore its own p16 is not obviously a cell that will respect a pharmacological copy.

The second argument is, as far as this beat can tell, unaddressed in the literature and is the sharpest open question this entry produces.

Facts and figures
mechanism classepigenetic repression of a tumour-suppressor locus feeding a cell-cycle checkpoint; contains a feedback loop
feedback arrowCDK6 is part of a transcription complex that INDUCES p16INK4a, kinase-independently - so the circuit closes on itself and an ATP-competitive inhibitor does not reproduce a CDK6 knockout
measured in index diseaseone row of six - the CDKN2A deletion frequency. Everything else is murine (MLL-AF9 or BCR-ABL) or human myeloid.
the unmeasured loopp16INK4a protein under a CDK4/6 inhibitor in a CDKN2A-intact KMT2A-rearranged cell. Not found in anything read this cycle.
nodes
[
  "gene-bmi1",
  "gene-cdkn2a",
  "gene-cdk6",
  "gene-cdk4",
  "gene-hoxa9"
]
forward arrows
[
  "BMI1 (PCGF4/canonical PRC1) represses CDKN2A/CDKN2B",
  "HOXA9 independently regulates the p16Ink4a/p19Arf locus and substitutes for BMI1 in murine MLL-AF9",
  "CDKN2A p16INK4a inhibits CDK4 and CDK6",
  "CDK4/CDK6 with a D-type cyclin phosphorylates RB1 and permits G1 exit"
]
state in infant kmt2a r
{
  "CDKN2A_deleted_pct": 8.5,
  "CDKN2A_deleted_n": "4/47",
  "comparator": "5/18 (28%) in infant ALL without KMT2A rearrangement, P=0.058",
  "implication": "the brake is present and epigenetically held down in roughly nine patients in ten, which is the opposite of the CDKN2A-null setting where the CDK4/6 class was developed"
}
depmap live 2026 09 20
{
  "CDKN2A": {
    "all_lines": 0.157,
    "leukaemia": 0.1309
  },
  "BMI1": {
    "all_lines": 0.0585,
    "leukaemia": -0.0103
  },
  "RB1": {
    "all_lines": 0.1801,
    "leukaemia": 0.1127
  },
  "CDK4": {
    "all_lines": -0.7371,
    "leukaemia": -0.4464
  },
  "CDK6": {
    "all_lines": -0.6321,
    "leukaemia": -1.3767
  },
  "CCND3": {
    "all_lines": -0.3459,
    "leukaemia": -1.0076
  },
  "n_leukaemia_lines": 57
}
How this could bite us back
  • Every intervention in this circuit ends at the same place, so a BMI1 agent and a CDK4/6 inhibitor are candidates for redundancy rather than synergy - and combining them doubles marrow toxicity for one mechanism.
  • Derepressing CDKN2A is a senescence programme, not a killing programme. Senescent blasts are alive and secretory, and the senescence-associated secretory phenotype can support neighbouring tumour cells.
  • The loop may defeat the drug. If CDK6's kinase-independent complex induces p16INK4a [E2], an ATP-competitive inhibitor removes the kinase and may also remove the induction, relieving the endogenous brake at the same time as the pharmacological one is applied.
  • The brake being intact in 91% of infant patients [E3] means the class is being proposed OUTSIDE the genotype it was developed for. A cell that already ignores its own p16 is not obviously a cell that will respect a drug that mimics it.
  • Five of the six links in this circuit have never been measured in infant B-lineage KMT2A::AFF1 human cells. The diagram is assembled from murine MLL-AF9, murine BCR-ABL and human myeloid systems, and the confidence of the whole is the confidence of its weakest transfer.
Still open
  1. Does p16INK4a protein RISE or FALL when a KMT2A-rearranged cell with an intact CDKN2A locus is treated with palbociclib or abemaciclib? The kinase-independent induction [E2] predicts the drug could relieve its own brake. One western blot answers it and nobody has published it.
  2. Does BMI1 loss derepress CDKN2A in a KMT2A-rearranged cell? The axis is assumed throughout this literature and the measurement in this disease was not found. If HOXA9 holds the locus down [E1], BMI1 loss alone may do nothing - which would be the cleanest possible confirmation of the substitution model.
  3. Are the ~8.5% of infant patients with CDKN2A deletion the ones who would respond to a CDK4/6 inhibitor, on the standard biomarker logic? They are also plausibly the higher-risk patients, and nobody has stratified a CDK4/6 result in this disease by CDKN2A status because no such result has been published.
  4. Is the CDKN2A repression here PRC1-mediated, PRC2-mediated, or DNA-methylation-mediated? This graph carries entries for all three arms and the locus's repressive mark in a KMT2A-rearranged blast was not established by anything read this cycle.
Every source behind this entry (5)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    In murine cells, MLL-AF9, which drives Hoxa7 and Hoxa9 expression, readily transforms Bmi1(-/-) cells while AML1-ETO and PLZF-RARalpha, which do not activate Hox, trigger senescence there; Hoxa9 regulates expression of the p16(Ink4a)/p19(ARF) locus as Bmi1 does and could overcome AML1-ETO-induced senescence; MLL-AF9 could not initiate leukaemia in Bmi1(-/-)Hoxa9(-/-) mice but either gene alone restored its ability to establish leukaemic stem cells.

    Functional crosstalk between Bmi1 and MLL/Hoxa9 axis in establishment of normal hematopoietic and leukemic stem cells · Smith LL, Yeung J, Zeisig BB, Popov N, Huijbers I, Barnes J, Wilson AJ, Taskesen E, Delwel R, Gil J, Van Lohuizen M, So CW · Cell Stem Cell 8:649-662 2011
    PMID 21624810DOI
    What this evidence cannot show

    Abstract read this cycle. MURINE MLL-AF9 MYELOID disease with germline knockouts; not infant B-lineage KMT2A::AFF1. The substitution is demonstrated at the p16Ink4a/p19Arf locus specifically and says nothing about BMI1's other polycomb targets.

    Evidence that argues against this (1)
    • A second 2011 study in the same fusion found Bmi1 indispensable for leukaemia in vivo and that deleting Ink4a and Arf only partially restored leukaemogenic capacity - so the circuit does not account for all of BMI1's requirement even in the system where it was measured. Yuan J et al. 2011
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    In a murine p185 BCR-ABL+ B-acute lymphoid leukaemia model, CDK6 is part of a transcription complex that induces expression of p16INK4a and VEGF-A, independently of CDK6's kinase activity; high CDK6 expression therefore suppresses proliferation by upregulating p16INK4a as an internal safeguard, and in the absence of p16INK4a CDK6 exerts its full tumour-promoting function - which the authors propose underlies a selection pressure to upregulate CDK6 and silence p16INK4a together.

    A kinase-independent function of CDK6 links the cell cycle to tumor angiogenesis · Kollmann K, Heller G, Schneckenleithner C, Warsch W, Scheicher R, Ott RG, Schafer M, Fajmann S, Schlederer M, Schiefer AI, Reichart U, Mayerhofer M, Hoeller C, Zochbauer-Muller S, Kerjaschki D, Bock C, Kenner L, Hoefler G, Freissmuth M, Green AR, Moriggl R, Busslinger M, Malumbres M, Sexl V · Cancer Cell 24:167-181 2013
    PMID 23948297DOI
    What this evidence cannot show

    Abstract read this cycle. MURINE BCR-ABL+ B-ALL, not KMT2A-rearranged. Umlauts in Schafer, Zochbauer-Muller are rendered as plain letters here; PubMed gives Schafer M and Zochbauer-Muller S with diacritics. The load-bearing half for this entry is the kinase-independent p16INK4a induction; the VEGF/angiogenesis half is not assumed to transfer to a marrow-resident leukaemia.

    Evidence that argues against this (1)
    • Two independent groups report that small-molecule CDK6 or CDK4/6 inhibition phenocopies CDK6 depletion in MLL-rearranged cells, which argues the catalytic arm carries real weight in this specific context whatever the scaffold does elsewhere. Placke T et al. 2014
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    CDKN2A is deleted in about 8.5% (4/47) of infant KMT2A-rearranged ALL against 28% (5/18) of infant ALL without a KMT2A rearrangement (P=0.058), so the p16INK4a/p14ARF locus is genomically intact in the large majority of infant KMT2A-rearranged patients and any repression of it is epigenetic rather than structural.

    The landscape of somatic mutations in infant MLL-rearranged acute lymphoblastic leukemias · Andersson AK, Ma J, Wang J, Chen X, Gedman AL, Dang J, Nakitandwe J, Holmfeldt L, Parker M, et al. · Nature Genetics 47:330-337 2015
    PMID 25730765DOI
    What this evidence cannot show

    The FREQUENCY is quoted from this graph's gene-cdkn2a entry and was not read from the paper by this beat; the PMID was seen in that sibling entry and then FETCHED FRESH from PubMed efetch on 2026-09-20, from which the title, DOI, venue and authors here were copied, because the brief forbids re-using a sibling entry's citation block. n=47 against n=18 with P=0.058 is underpowered, and no clonality data are reported in any study gene-cdkn2a read - so '91% intact' describes the diagnostic sample, not necessarily the relapsing clone.

    Evidence that argues against this (1)
    • In NON-infant KMT2A-rearranged disease the split is by lineage: CDKN2A/B deletions in 3 of 9 (33%) ALL cases and none of the AML cases, per the same source as recorded in gene-cdkn2a - so 'intact' is an infant-specific statement and does not generalise across ages. Andersson AK et al. 2015
  4. E4established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    Fetched live on 2026-09-20, DepMap Chronos mean gene effects across all lines and across 57 leukaemia-annotated lines are: CDKN2A +0.1570 / +0.1309; BMI1 +0.0585 / -0.0103; RB1 +0.1801 / +0.1127; CDK4 -0.7371 / -0.4464; CDK6 -0.6321 / -1.3767; CCND3 -0.3459 / -1.0076. Of the six nodes in this circuit, only CDK6 and its cyclin CCND3 are fitness requirements of leukaemia lines, and the three tumour-suppressor or polycomb nodes are neutral to favourable on knockout.

    DepMap CRISPR (Chronos) gene effect for CDKN2A, BMI1, RB1, CDK4, CDK6 and CCND3, retrieved via the Open Targets Platform GraphQL API, depMapEssentiality.screens · Broad Institute DepMap; Open Targets Platform; query run by this project 2026-09-20 · Database query 2026
    What this evidence cannot show

    A database fact opposable on scope. A positive gene effect on a tumour suppressor is close to uninterpretable as biology: many lines already carry CDKN2A or RB1 deletions, so the knockout has nothing to remove, and a CRISPR cut at a deleted locus produces its own artefacts. The leukaemia subset is a coarse text match on diseaseFromSource, not curated lineage, and is not conditioned on fusion status. No infant-derived line in the panel. 21-day fitness only - BMI1's published phenotype is self-renewal and is invisible here.

    Evidence that argues against this (1)
    • The CDKN2A gene-effect number in particular should not be read as 'losing the brake helps': DepMap scores at a locus that is homozygously deleted in a large fraction of cancer lines are dominated by copy-number and cutting artefacts rather than by biology, and this beat did not check the copy-number distribution. Broad Institute DepMap; Open Targets Platform 2026
  5. E5established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    Bmi-1 regulates the proliferative activity of both normal and leukaemic stem and progenitor cells in mice, with Bmi-1-null leukaemic cells undergoing proliferation arrest, differentiation and apoptosis and failing to transplant the leukaemia; and the Ink4a/Arf locus is identified in the polycomb literature as the major target through which Bmi1 acts in leukaemic stem cells.

    Bmi-1 determines the proliferative capacity of normal and leukaemic stem cells · Lessard J, Sauvageau G · Nature 423:255-260 2003
    PMID 12714970DOI
    What this evidence cannot show

    Abstract read this cycle. The Ink4a/Arf half of this claim is NOT in the abstract read; it is the field's standing position as restated in the 2011 papers cited elsewhere in this entry (PMID 21527932: 'one of their major targets in LSCs is known to be the Ink4a/Arf tumor suppressor gene locus'). The leukaemia model is Hoxa9+Meis1a-driven, which is a poor neutral test of an axis whose whole question is whether Hoxa9 substitutes for Bmi1.

    Evidence that argues against this (1)
    • Deleting both Ink4a and Arf only PARTIALLY restored the leukaemogenic capacity of Bmi1-null LSCs, so the Ink4a/Arf locus is not the whole of BMI1's effect even in a murine MLL-AF9 system. Yuan J et al. 2011
Connected to