Oncofusion

start / graph / target-cdk9-inhibition-kmt2a-r

Targetmoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q6b-therapeutics2

CDK9 inhibition in KMT2A-rearranged infant leukemia — the most direct hit on the hijacked machinery, and the narrowest window

also called P-TEFb inhibition · enitociclib · VIP152 · alvocidib · AZD4573 · flavopiridol

CDK9 is the engine the fusion protein steals to force cancer genes to be read continuously. Blocking it shuts down the two proteins this leukemia most depends on, MYC and MCL-1, but CDK9 is needed by every cell in the body, which is the problem.

The same thing, for a scientist

CDK9, with cyclin T, forms P-TEFb, the kinase that releases paused RNA polymerase II into productive elongation. KMT2A fusions recruit P-TEFb through the super elongation complex, making CDK9 the most mechanistically direct target in the fusion's machinery. The clinical-stage selective inhibitor enitociclib (VIP152) inhibits growth of KMT2A-rearranged infant leukemia cells with on-target loss of MYC and MCL-1, overcomes the bone-marrow-microenvironment growth advantage, synergises with menin inhibitors to deplete HOXA9 protein, and potentiates venetoclax in venetoclax-resistant KMT2A-rearranged cells. CDK9 is DepMap common-essential, which is the reason this class has repeatedly failed on toxicity despite twenty-five years of attempts.

Go deeper
The specific infant-leukemia data

Enitociclib, a selective CDK9 inhibitor in clinical development, tested in KMT2A-rearranged infant leukemia 1:

  • growth inhibition with on-target decreases in MYC and MCL-1 protein;
  • reduced the growth advantage conferred by the bone marrow microenvironment — a notable result because stromal protection is one of the reasons in vitro sensitivity fails to translate;
  • synergised with menin inhibitors, producing an effective decrease in HOXA9 protein;
  • potentiated venetoclax cytotoxicity in relatively venetoclax-resistant KMT2A-rearranged cells, consistent with MCL-1 being the escape route from BCL-2 inhibition.

Those three combination properties — microenvironment, menin, venetoclax — make CDK9 unusually well positioned as the third agent in a menin/BCL-2 backbone. The mechanism is coherent: menin inhibition evicts the fusion from chromatin, venetoclax lowers the apoptotic threshold, and CDK9 inhibition removes the MCL-1 buffer and the MYC output simultaneously.

Why this class keeps failing, and what would have to be different

CDK9 is DepMap common-essential: mean Chronos below -0.8, depleted in over 92% of cell lines 2. Global transcription is not a tissue-selective process. Flavopiridol/alvocidib and successive CDK9 inhibitors have produced responses in AML and have repeatedly foundered on myelosuppression, tumour lysis and gastrointestinal toxicity.

The arguments for trying again are specific rather than general:

  1. Short-exposure pharmacology. MCL-1 and MYC have half-lives measured in tens of minutes. A brief, deep, intermittent CDK9 inhibition may collapse them while normal tissue, which tolerates transient transcriptional pause, recovers. Newer agents (enitociclib, AZD4573) are designed around exactly this.
  2. Combination allows lower exposure. If the role is to remove the MCL-1 buffer so venetoclax can work, the required CDK9 inhibition is partial and brief, not sustained.
  3. The dependency is real and non-redundant. Unlike DOT1L, whose enzymatic function the leukemia learned to live without, transcriptional elongation is not something a cell can rewire around.

The argument against is the one that should hold: the same three arguments were made for every previous CDK9 inhibitor. No CDK9 inhibitor is approved for any indication. This node asserts no clinical response rate in KMT2A-rearranged disease because we could not verify one from a primary source.

Facts and figures
modalitysmall-molecule kinase inhibitors; short-exposure intermittent dosing is the current design strategy
clinical status in kmt2a rNo CDK9 inhibitor is approved for any indication. No verified clinical response rate in KMT2A-rearranged leukemia was found from a primary source.
depmap common essentialtrue
ccnt1 statusCCNT1 was struck as a standalone candidate in cycle 22 (see concept-phase2-go-no-go-register and ana-target-list-rewrite-audit) on co-dependency grounds. This entry's target is CDK9 and is unaffected. The clinical argument offered alongside that strike — 7 of 9 treatment-related deaths in the alvocidib arm of PMID 30370956 — is recorded on E4 and explicitly rejected as a reason to stop CDK9 work: that arm had the best response rate of three, was seven years older than its comparators, used a pan-CDK drug with two cytotoxics, and its authors recommended further development.
alvocidib trial discordanceTwo randomised alvocidib trials disagree about toxicity. Newly diagnosed, n=165 (PMID 26022709): higher CR than 7+3, explicitly without increased toxicity, no survival difference. Relapsed/refractory, n=90 across three arms (PMID 30370956): best response rate but 7 of 9 treatment-related deaths, in an arm seven years older. Neither supports a general verdict on P-TEFb inhibition. The durable objection to the class is that extra remissions did not become extra survivors.
pediatric eligiblefalse
target genes
[
  "CDK9",
  "CCNT1"
]
agents
[
  "enitociclib (VIP152)",
  "AZD4573",
  "alvocidib (flavopiridol)"
]
downstream effectors
[
  "MYC",
  "MCL1"
]
combination evidence
[
  "synergy with menin inhibitors (HOXA9 protein depletion)",
  "potentiation of venetoclax in venetoclax-resistant KMT2A-rearranged cells",
  "abrogation of bone-marrow-microenvironment-mediated growth advantage"
]
How this could bite us back
  • Essential-function toxicity: CDK9 drives global transcriptional elongation; the on-target effect in normal proliferating tissue is the dose-limiting toxicity and this has killed the class repeatedly.
  • Tumour lysis syndrome: rapid MCL-1 and MYC collapse in a high-burden infant leukemia is exactly the setting where flavopiridol caused fatal tumour lysis in adults.
  • Myelosuppression on top of an already-failing marrow, compounding the infection risk that dominates infant ALL mortality.
  • Compensation: CDK12/CDK13 and other transcriptional CDKs provide partial redundancy, and cross-inhibition of them widens toxicity without necessarily widening efficacy.
  • Combination hazard: adding CDK9 inhibition to menin plus venetoclax stacks three agents that all lower the apoptotic threshold, with no paediatric dose-finding anywhere in that space.
  • Treatment-related mortality in the relapsed/refractory setting is now on the record for this class (E4), even after the confounding is discounted. The FLAM arm's deaths concentrated in older patients with a heavier prior treatment burden, and an infant relapsing after menin plus chemotherapy is the paediatric analogue of that patient: a fragile marrow, prior anthracycline exposure, and the highest disease burden. The confounders that make the number uninterpretable as a P-TEFb effect do not make it irrelevant as a regimen effect.
  • This entry has never had a registered go/no-go gate table, despite being the most mechanistically direct target in the fusion's machinery and the one with clinical-stage compounds. A programme with no STOP criterion is the failure mode the register was built to prevent.
Still open
  1. Does short-exposure CDK9 inhibition genuinely spare normal haematopoiesis, or does it simply move the toxicity? The pharmacodynamic argument has never been tested with a normal-HSC competitive repopulation readout.
  2. Is the menin-plus-CDK9 synergy additive transcriptional suppression or something specific to HOXA9 protein turnover? The distinction determines whether CDK9 is interchangeable with any other elongation inhibitor.
  3. Would CDK9 inhibition pre-empt or accelerate adaptive menin resistance? It removes MYC, which is the named bypass in PRC1.1-deficient resistance, so there is a specific prediction to test.
  4. Can any CDK9 inhibitor be given to an infant? There are no paediatric pharmacokinetic data for enitociclib or AZD4573 and tumour lysis risk in a hyperleukocytotic infant is severe.
Every source behind this entry (5)
  1. E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    The clinical-stage CDK9 inhibitor enitociclib inhibits growth of KMT2A-rearranged leukemia cells with on-target decreases in MYC and MCL-1 protein, reduces the growth advantage provided by the bone marrow microenvironment, synergises with menin inhibitors to decrease HOXA9 protein in KMT2A-rearranged infant leukemia cells, and potentiates venetoclax cytotoxicity in relatively venetoclax-resistant KMT2A-rearranged cells.

    Targeting CDK9 inhibits the growth of KMT2A-rearranged infant leukemia and demonstrates synergy with menin inhibition · See PubMed record PMID 40979068 for the full author list (not transcribed here) · Blood Neoplasia 2025
    PMID 40979068DOI
    What this evidence cannot show

    In vitro only — the authors explicitly describe 'measurable in vitro antitumor activity' and propose future clinical trials. No in vivo xenograft efficacy is claimed in the abstract. Synergy metrics in ex vivo leukemia culture are assay-duration sensitive. Enitociclib has no paediatric data and no leukemia approval.

    Evidence that argues against this (1)
    • Clinical CDK9 inhibition has already produced the pattern this preclinical work predicts, and then stalled at exactly the step that matters. In a 165-patient randomised phase 2 in newly diagnosed intermediate/adverse-risk AML, the CDK9 inhibitor flavopiridol (alvocidib) in the FLAM regimen gave a significantly higher complete remission rate than 7+3 (70% vs 46%, p=0.003) without increased toxicity, yet there were no significant differences in overall survival or event-free survival between the arms. More cells killed up front did not become more patients alive - the outcome any enitociclib combination has to beat. Zeidner JF et al. 2015
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    CDK9 is a DepMap common-essential gene with mean Chronos gene effect below -0.8 and depletion in more than 92% of cell lines, unlike MEN1.

    What this evidence cannot show

    Cell-line essentiality models normal-tissue tolerance only loosely, and short intermittent pharmacological inhibition is not equivalent to constitutive genetic knockout — which is precisely the argument the newer CDK9 inhibitors are built on. Values were computed from an API response by another agent in this project rather than read from a DepMap release table. No counter-argument is recorded and none is available in kind: this is a CRISPR-screen statistic from a database rather than a contestable inference, and no published study disputes the Chronos value. What is contestable is the therapeutic reading of it, and there the counter-argument is empirical - alvocidib was given to hundreds of AML patients at doses producing real antileukaemic activity without a prohibitive on-target window (see the counter-evidence on E1), so in-vitro common-essentiality did not translate into an unusable clinical window; it translated into no survival benefit.

    Evidence that argues against this (1)
    • Pan-line CRISPR essentiality did not predict what happened when CDK9 was inhibited in patients, in either direction. In the phase 1 of the oral CDK9 inhibitor voruciclib (40 patients, 21 with AML, median 3 prior lines), dose escalation on an intermittent days 1-14 schedule reached 200 mg with NO dose-limiting toxicities at a plasma concentration sufficient for target inhibition, and the commonest adverse events were diarrhoea (30%), nausea (25%) and anaemia (22%) — not the catastrophic on-target toxicity a 'common essential in more than 92% of cell lines' score suggests. Target engagement was documented in patients (reduced phosphorylation of RNA polymerase II, decreased MCL1 mRNA, downregulated MYC and NF-kB gene sets). Chronos scores index what a gene does to a proliferating cell line over two weeks, not what a drug does to a patient on an intermittent schedule. Davids MS et al. 2025
  3. E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    KMT2A fusions drive leukemia in part by recruiting the super elongation complex containing P-TEFb (CDK9/cyclin T) to target loci, deregulating transcriptional elongation and upregulating HOXA9.

    Targeting CDK9 inhibits the growth of KMT2A-rearranged infant leukemia and demonstrates synergy with menin inhibition. · See PubMed record PMID 40979068 for the full author list (not transcribed here) · Blood Neoplasia 2025
    PMID 40979068DOI
    What this evidence cannot show

    This mechanism is textbook-level and supported by a large primary literature elsewhere in this graph (see mechanism-sec-elongation-hijack and protein-ptefb-cdk9-ccnt1); it is cited here only via this paper's framing rather than from the original elongation-complex papers.

    Evidence that argues against this (1)
    • The SEC/P-TEFb account predicts that inhibiting CDK9 should collapse the fusion's elongation program and with it the leukaemia; the one recent clinical study with matched pharmacodynamics shows the first half happening without the second. Voruciclib reduced RNA polymerase II phosphorylation, lowered MCL1 messenger RNA and downregulated MYC and NF-kB transcriptional gene sets in treated patients — and among 21 patients with AML, ONE achieved a morphologic leukaemia-free state and two had stable disease, with no complete remissions. Demonstrated on-target transcriptional effect with essentially no clinical response bounds how much therapeutic weight the elongation mechanism can carry on its own. Davids MS et al. 2025
  4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    A second randomised alvocidib trial, in relapsed/refractory rather than newly diagnosed AML, reports the toxicity signal the frontline trial did not: of 90 patients randomised across three regimens, the alvocidib-containing FLAM arm had the highest overall response (28% vs 14% and 16%) and 7 of the trial's 9 treatment-related deaths, 4 of those in elderly patients, with a median age of 62 in that arm against 55 in the comparators.

    A randomized trial of three novel regimens for recurrent acute myeloid leukemia demonstrates the continuing challenge of treating this difficult disease · See PubMed record PMID 30370956 for the full author list (not transcribed here) · American Journal of Hematology 2018
    PMID 30370956DOI
    What this evidence cannot show

    Abstract read, not full text. Added in cycle 22 because a counter-evidence sweep had attached this trial to CCNT1 as evidence that P-TEFb inhibition in AML produces toxicity rather than selectivity, and this entry did not carry it. It is added here as a record of the data, NOT as an endorsement of that reading, which ana-target-list-rewrite-audit rejects. Four reasons the mortality figure cannot be attributed to P-TEFb: the arms are 35, 36 and 19 patients under a Simon two-stage design per arm and were never powered for a head-to-head safety comparison; the FLAM arm was significantly older, which the abstract states and does not adjust for; alvocidib is a multi-CDK inhibitor, not a selective CDK9 agent; and it was given with cytarabine and mitoxantrone, so no event is attributable to CDK9 inhibition alone. The authors' own conclusion is that FLAM 'had an encouraging response rate and should be considered for further clinical development but should be used with caution in elderly patients.'

    Evidence that argues against this (1)
    • The two alvocidib trials disagree about toxicity and the graph should hold both rather than pick one. The larger, newly diagnosed trial already cited on E1 (165 patients) reports a significantly higher complete remission rate than 7+3 explicitly WITHOUT increased toxicity. Same drug, same regimen backbone, opposite toxicity reading — which is what one expects when the difference between the populations is relapsed/refractory versus newly diagnosed and a seven-year median-age gap. Neither trial supports a general verdict on P-TEFb inhibition, and the durable objection to the class remains the one on E1: extra remissions did not become extra survivors. Zeidner JF et al. 2015
  5. E5moderate confidence 3 of 5: One strong study, or several smaller consistent ones.database

    CCNT1, the cyclin partner that forms P-TEFb with CDK9, was struck as a standalone candidate in cycle 22 on co-dependency grounds: its MEN1 correlation of r = 0.349 across 54 DepMap leukemia lines fails Holm correction in the 16-gene panel (Holm p = 0.146) and falls below every gene that survives the 2,743-gene variance-matched deep-pool null. Unlike CDK9, CCNT1 is not DepMap common-essential (mean Chronos -0.206, 8.4% of lines depleted).

    DepMap CRISPR (Chronos) gene effect and MEN1 co-dependency for CCNT1 and CDK9, via the Open Targets Platform GraphQL API; deep-pool null in ana-depmap-null-deep-pool · Broad Institute DepMap; Open Targets Platform; analyses run by this project · Open Targets and derived analysis 2026
    What this evidence cannot show

    This project's own unreviewed computation on 54 leukemia lines with an uncontrolled lineage confounder, and two earlier versions of the same analysis reached different conclusions using a criterion since shown to be unstable. DepMap release version unverified. The striking of CCNT1 does not touch this entry's target, which is CDK9: the two are halves of one enzyme but they are not interchangeable as targets, because the chemistry, the structures and every clinical-stage compound are on the kinase.

    Evidence that argues against this (1)
    • A DepMap co-dependency correlation is the wrong instrument for striking CCNT1, because the paralogue it would be buffered by is present in the very complex at issue. Mass-spectrometric purification of the AF9 C-terminal transforming domain from myeloblastic M1 cells identified an 'elongation assisting proteins' (EAP) complex containing Enl, Af4, Laf4, Af5q31, Ell and Af10 together with the RNA polymerase II CTD kinase Cdk9 and CYCLIN T1/T2 — i.e. P-TEFb in this complex is assembled with either cyclin T paralogue. A single-gene CRISPR co-dependency signal cannot distinguish 'CCNT1 is not required' from 'CCNT1 loss is compensated by CCNT2', and a failed Holm correction on r = 0.349 across 54 lines is exactly the pattern paralogue buffering produces. The strike should be recorded as untested rather than as evidence against CCNT1. Monroe SC et al. 2011
Connected to