Oncofusion

start / graph / target-cd33-cd123-myeloid-directed-kmt2a-r

Targetmoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q6b-therapeutics2disputed

Myeloid-directed antigens (CD33, CD123) — the only immunotherapy axis that survives lineage switch

also called gemtuzumab ozogamicin · Mylotarg · CD33 ADC · CD123 · IL3RA · tagraxofusp

When KMT2A-rearranged leukemia escapes CD19-targeted therapy by turning into a myeloid leukemia, it usually starts wearing myeloid markers like CD33 instead. Drugs aimed at those markers are the only immune option that follows the leukemia where it goes.

The same thing, for a scientist

CD33 and CD123 are myeloid surface antigens with approved or clinical-stage targeting agents. Their relevance in KMT2A-rearranged disease is threefold: KMT2A-rearranged AML is a subgroup in which gemtuzumab ozogamicin improved outcome in COG AAML0531; KMT2A-rearranged B-ALL frequently co-expresses myeloid antigens at diagnosis; and pinometostat-resistant KMT2A::AFF1 B-ALL cells upregulated CD33 and LILRB4 as part of their adaptive myeloid drift. Together these make myeloid-antigen targeting the only immunotherapy axis that does not collapse under lineage switch. Gemtuzumab plus ziftomenib plus venetoclax is already in a paediatric phase 1 (NCT06448013).

Go deeper
What is established, and how firmly

CD33 / gemtuzumab ozogamicin. The Children's Oncology Group states that children with KMT2A-rearranged AML benefited from the addition of gemtuzumab ozogamicin, an anti-CD33 antibody-drug conjugate, in the AAML0531 trial (NCT00372593) 1. This graph records that as a moderate-confidence claim because the source we verified is the COG's 2023 research blueprint, a review, rather than the primary AAML0531 subgroup report, which we could not locate in PubMed and therefore do not cite. The effect size and confidence interval are not stated here for that reason.

CD123 / IL3RA. We found no verified primary source for CD123-directed therapy in KMT2A-rearranged or infant leukemia in this sweep. CD123 is widely discussed in this context and tagraxofusp and CD123 CAR-T constructs exist, but this node makes no claim about them. Anything asserted about CD123 here would be conference material or extrapolation, and per this graph's rules that is left out.

The strategic argument, which is stronger than the current evidence

Three independent observations point the same way.

  1. KMT2A-rearranged B-ALL is myeloid-adjacent at baseline. Co-expression of myeloid antigens is part of the diagnostic phenotype, and a substantial fraction are formally mixed phenotype acute leukemia — which is why Interfant-21 and NCT06317662 explicitly enrol MPAL alongside ALL.
  2. Therapy pushes it further myeloid. Pinometostat-resistant KMT2A::AFF1 B-ALL cells upregulated CD33 and LILRB4/CD85k while selectively losing other fusion targets 2. This is epigenetic-therapy pressure producing myeloid drift in the absence of any immune selection.
  3. Immune pressure completes the switch. CD19-directed therapy in this genotype selects frank myeloid lineage switch, a mechanism covered separately in this graph and demonstrated in a post-tisagenlecleucel ALL-to-AML patient-derived xenograft pairing 3.

Put together: every therapy class the field is deploying pushes this leukemia toward a myeloid phenotype, and the only immune agents that follow it there are CD33- and CD123-directed. Adding a myeloid-directed agent up front, rather than after the switch has happened, is the strategically obvious move and nobody is testing it in infants.

The nearest thing in the clinic is NCT06448013 (MD Anderson, phase 1, recruiting, n=22): ziftomenib with venetoclax and gemtuzumab in paediatric AML 4. That is a menin inhibitor plus a BCL-2 inhibitor plus a CD33 ADC — the combination this reasoning predicts — but it is in AML, not in KMT2A-rearranged infant ALL where the lineage-switch risk is greatest.

Correction: the overall survival benefit was not significant

This entry originally rested on a review rather than the primary trial report. The primary report of COG AAML0531 shows gemtuzumab ozogamicin improved event-free survival, disease-free survival and relapse risk in KMT2A-rearranged AML, but the overall survival difference did not reach significance (63% vs 53%, p = 0.054) 0.

A second qualification matters more for a platform argument: the benefit appears restricted by germline genotype. Relapse fell from 49% to 26% in children homozygous for one CD33 splicing variant, and was unchanged in the other genotypes, whose minor allele frequency is near 30%. On that evidence roughly a third of children may derive no benefit, and a trial that does not genotype cannot tell which.

Correction: CD33 and CD123 point in OPPOSITE directions, and this entry bundles them

This entry treats CD33- and CD123-directed therapy as one strategy. Two primary records, both fetched and read back by the coordinator on 2026-09-20, show the two antigens do not behave the same way in this disease - and the difference is lineage, which is exactly the axis this entry reasons across.

recorddiseasewhat it measuresdirection for KMT2A-r
PMID 34855461 (COG AAML1031, n=1,040)childhood AMLCD123 surface expression by quartilehighest quartile is enriched for KMT2A rearrangement and FLT3-ITD
PMID 38468826B-ALLmarker MFI, KMT2A-r vs notKMT2A-r has significantly LOWER CD123 MFI, alongside lower CD34, CD22 and TdT

So the CD123 argument in this entry was imported from AML into B-lineage disease, where the marker moves the other way. The strategic claim above - that adding a myeloid-directed agent up front is the obvious move in infants - survives for CD33 and is actively weakened for CD123, because the B-lineage measurement says there is less CD123 on these blasts, not more. The entry's own admission that it found "no verified primary source for CD123-directed therapy in KMT2A-rearranged or infant leukemia" was correct and is now sharper: the nearest primary evidence points away.

This is also a standing-rule violation on our side, not the literature's. The project rule is one entry per thing; never a summary entry covering several. Bundling two antigens into one target entry is what allowed an AML-derived direction to be carried silently onto a B-lineage target, because there was no CD123 entry in which the contradiction would have had to be faced. gene-il3ra now exists and carries the CD123 biology; this entry should be split, and until it is, read its CD123 claims against that entry.

Facts and figures
modalityantibody-drug conjugate (gemtuzumab ozogamicin, CD33); CD123-directed agents include a diphtheria-toxin fusion and CAR T constructs
status cd33Gemtuzumab ozogamicin approved in paediatric AML; benefit in the KMT2A-rearranged subgroup reported by COG but not verified here from a primary source
status cd123No verified primary source identified in this sweep for CD123-directed therapy in KMT2A-rearranged or infant leukemia. No claim is made.
pediatric eligibletrue
strategic rationalethe only immunotherapy axis that remains valid after myeloid lineage switch, the dominant genotype-specific escape from CD19-directed therapy
target genes
[
  "CD33",
  "IL3RA"
]
trials
[
  "NCT00372593",
  "NCT06448013"
]
supporting observations
[
  "KMT2A-rearranged B-ALL frequently co-expresses myeloid antigens and a substantial fraction is MPAL",
  "DOT1L-inhibitor-resistant KMT2A::AFF1 B-ALL upregulates CD33 and LILRB4",
  "CD19-directed therapy selects frank myeloid lineage switch in this genotype"
]
How this could bite us back
  • Essential-function toxicity: CD33 and CD123 are both expressed on normal myeloid progenitors, so on-target myelosuppression is unavoidable; CD123 is additionally expressed on endothelium, which is the origin of the capillary leak syndrome seen with CD123-directed agents.
  • Hepatic sinusoidal obstruction syndrome is a calicheamicin class toxicity shared with inotuzumab, and stacking CD22 then CD33 conjugates in one child's course compounds it.
  • Antigen escape in the other direction: a leukemia under simultaneous CD19 and CD33 pressure has been shown in principle to be able to drop either, and dual pressure may simply select a doubly-negative clone.
  • Evidence risk: the CD33 benefit claim in KMT2A-rearranged AML rests here on a review, and the CD123 story rests on nothing we could verify. This is exactly the kind of area where enthusiasm outruns primary data.
  • Late effects: prolonged myeloid progenitor depletion in an infant, on top of transplant conditioning, has an unquantified effect on lifelong haematopoietic reserve.
  • The overall survival benefit is not statistically significant, and the relapse benefit appears confined to one germline genotype, so a CD33-directed strategy may help far fewer children than the headline suggests.
  • Lineage-imported rationale: the CD123 case in this entry rests on a childhood AML cohort, and the only B-lineage measurement we could find says KMT2A-rearranged blasts carry LESS CD123, not more. Anyone citing this entry to justify a CD123-directed agent in infant B-ALL is carrying an AML result across a lineage boundary the data does not support crossing.
Still open
  1. What is the CD33 and CD123 surface density on KMT2A-rearranged infant B-ALL blasts at diagnosis, and does it rise under menin-inhibitor or blinatumomab pressure? This is a flow cytometry experiment on existing samples and it would directly justify or kill an up-front myeloid-directed strategy.
  2. Should a myeloid-directed agent be given BEFORE lineage switch rather than after? Every current trial adds it only in AML or at relapse, which means the field is always one step behind the escape it has itself caused.
  3. Is there any verified CD123 data in KMT2A-rearranged leukemia? The field talks about CD123 constantly and we could not verify a primary source; that gap should be closed or the topic dropped.
  4. Does gemtuzumab's benefit in KMT2A-rearranged AML extend to KMT2A-rearranged MPAL and lineage-switched relapses, which are the cases that most need it?
Every source behind this entry (6)
  1. E0strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    In the phase 3 COG AAML0531 trial, gemtuzumab ozogamicin improved 5-year event-free survival in KMT2A-rearranged AML (48% vs 29%, p=0.003), disease-free survival (57% vs 33%, p=0.002) and relapse risk (40% vs 66%, p=0.001), but the overall survival difference was NOT statistically significant (63% vs 53%, p=0.054).

    PMID 34048275DOI
    What this evidence cannot show

    215 of 1,022 enrolled patients had KMT2A-rearranged AML, so this is a subgroup of a trial not designed around it. Event-free survival benefit without a significant overall survival benefit is a weaker basis for changing practice in an infant.

    Evidence that argues against this (1)
    • The CD33 benefit appears restricted by germline genotype: relapse fell from 49% to 26% in rs12459419 CC patients but was unchanged in CT/TT (39% vs 40%, p=0.85), and the minor allele frequency is around 30%, so roughly a third of children may derive no benefit. Lamba JK et al. 2017
  2. E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.review

    Children with KMT2A-rearranged AML and/or FLT3-ITD benefited from the addition of gemtuzumab ozogamicin, an anti-CD33 antibody-drug conjugate, in the Children's Oncology Group AAML0531 clinical trial (NCT00372593).

    Children's Oncology Group's 2023 blueprint for research: Myeloid neoplasms · Cooper TM, Alonzo TA, Tasian SK, et al. · Pediatric Blood & Cancer 2023
    PMID 37480164DOI
    What this evidence cannot show

    This is a REVIEW used as a map, not a primary report. We could not locate the primary AAML0531 KMT2A subgroup publication in PubMed during this sweep, so no effect size, hazard ratio or confidence interval is asserted. Subgroup benefit reported in a review is one of the commonest routes by which unverified numbers enter a knowledge base, and this node deliberately stops short of quoting any. The claim also concerns AML, not KMT2A-rearranged infant ALL.

    Evidence that argues against this (3)
    • The primary AAML0531 analysis is weaker than the word 'benefited' suggests, and this entry cites a blueprint review rather than that analysis. In the 215 KMT2A-rearranged patients of AAML0531, gemtuzumab improved 5-year EFS (48% vs 29%, p=.003) and reduced relapse risk (40% vs 66%, p=.001), but overall survival was NOT significantly different (63% vs 53%, p=.054). The benefit is a relapse-risk benefit that did not reach significance for survival in the KMT2A-rearranged subset. Pollard JA et al. 2021
    • CD33-directed benefit is germline-restricted, which undercuts CD33 as a universal fallback antigen after lineage switch. In the same AAML0531 randomisation (816 genotyped patients), those with the CD33 rs12459419 CC genotype had lower relapse risk with gemtuzumab (26% vs 49%, p<.001) and better DFS (65% vs 46%, p=.004), whereas in CT/TT patients gemtuzumab made no difference at all (relapse 39% vs 40%, p=.85). The T allele (30% allele frequency) drives an alternatively spliced D2-CD33 isoform lacking the IgV antibody-binding domain, so a large minority of patients have no expected benefit and are also under-detected by standard CD33 immunophenotyping panels. Lamba JK et al. 2017
    • In adults, individual-patient-data meta-analysis of five randomised trials (3325 patients) found the gemtuzumab survival benefit was confined to favourable (20.7% absolute at 6 years, OR 0.47, p=0.0006) and intermediate cytogenetics (5.7%, OR 0.84, p=0.005), with NO benefit in adverse cytogenetics (2.2%, OR 0.99, 95% CI 0.83-1.18, p=0.9). Most KMT2A partners other than t(9;11) are classified adverse, so the adult evidence base argues against CD33-directed therapy in precisely the KMT2A-rearranged subgroups that do worst. Hills RK et al. 2014
  3. E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    Paediatric KMT2A::AFF1 B-ALL cells with acquired resistance to the DOT1L inhibitor pinometostat upregulated myeloid-associated genes including CD33 and LILRB4/CD85k while selectively losing expression of other KMT2A-fusion target genes.

    Modelling acquired resistance to DOT1L inhibition exhibits the adaptive potential of KMT2A-rearranged acute lymphoblastic leukemia · Schneider P, Crump NT, Arentsen-Peters STCJM, et al. · Experimental Hematology & Oncology 2023
    PMID 37740239DOI
    What this evidence cannot show

    A single in vitro cell-line model of acquired drug resistance; transcript-level upregulation of CD33 is not the same as a therapeutically targetable surface density, which was not measured. Whether the same drift occurs under menin-inhibitor pressure is unknown.

    Evidence that argues against this (1)
    • Upregulated CD33 on drug-adapted cells does not license the expectation that a CD33-directed agent converts that into survival. The largest randomised test of CD33 targeting in paediatric KMT2A-rearranged AML — 215 KMT2A-r patients among 1,022 enrolled on Children's Oncology Group AAML0531 — found that gemtuzumab ozogamicin improved 5-year EVENT-FREE survival (48% with versus 29% without, P=.003), lowered relapse risk (40% versus 66%, P=.001) and improved disease-free survival (57% versus 33%, P=.002), but OVERALL SURVIVAL WAS NOT SIGNIFICANTLY DIFFERENT (63% versus 53%, P=.054). A CD33-directed agent in this disease buys relapse-free time that has not been shown to convert into lives saved, which is the honest ceiling for a myeloid-directed strategy built on antigen upregulation. Pollard JA et al. 2021
  4. E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    A post-tisagenlecleucel ALL-to-AML lineage switch patient-derived xenograft model pairing exists and was used to demonstrate activity of FLT3-directed CAR T cells against KMT2A-rearranged leukemia after CD19-directed therapy failure.

    PMID 35950535DOI
    What this evidence cannot show

    A single patient-derived model pair, used to demonstrate a CAR construct rather than to characterise lineage switch frequency. It establishes that the phenomenon is real and modellable, not how often it happens.

    Evidence that argues against this (1)
    • The lineage-switch pairing is being used as a model of trans-differentiation under CD19 pressure, and the genomic evidence says it is clonal selection instead. Whole-genome sequencing of serial leukaemias in a child treated with anti-CD19 cellular and antibody therapy established that the subsequent AML derived from the initial ALL and carried distinct driver mutations DETECTABLE BEFORE the switch emerged, following extensive prior clonal diversification. If the myeloid clone pre-exists, then a myeloid-directed agent given after CD19 failure is treating a lineage that was selectable all along, and the therapeutic question becomes whether to target it pre-emptively rather than in rescue — a different trial design from the one this model motivates. Coorens THH et al. 2023
  5. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.trial_registry

    NCT06448013 is a recruiting phase 1 trial at MD Anderson (n=22) combining the menin inhibitor ziftomenib with venetoclax and gemtuzumab in paediatric patients with acute myeloid leukemia.

    NCT06448013
    What this evidence cannot show

    Registry record only; single site, opened March 2025, no results. The population is paediatric AML, not KMT2A-rearranged infant ALL, and 22 patients will not resolve a three-drug combination question.

    Evidence that argues against this (1)
    • This trial excludes the age group the entry is about and cannot estimate efficacy in any case. Raw ClinicalTrials.gov v2 API, retrieved 2026-09-19: NCT06448013 is a phase 1 study with ESTIMATED enrolment 22, MINIMUM AGE 3 YEARS and maximum age 21 years, whose SOLE primary outcome is 'Safety and adverse events' over an average of one year, with primary completion 2028-12-31 and study completion 2030-12-31. 'Paediatric' in this record means 3 to 21 years; infants with KMT2A-rearranged disease are not eligible, and a 22-patient safety study will not produce an efficacy estimate for the ziftomenib/venetoclax/gemtuzumab triplet at any age. M.D. Anderson Cancer Center 2025
  6. E-cd123-lineage-inversionmoderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    CD123 expression moves in opposite directions with KMT2A rearrangement depending on lineage: in childhood AML the highest CD123 quartile shows a higher prevalence of KMT2A rearrangements and FLT3-ITD, whereas in B-lineage ALL, KMT2A-rearranged cases show significantly lower CD123 mean fluorescence intensity than non-rearranged cases.

    PMID 34855461DOI
    What this evidence cannot show

    Both abstracts fetched and read back on 2026-09-20; neither full text read. **Moderate, and the limit is that this is a comparison across two studies, two diseases and two measurement conventions** - AAML1031 bins a quartile of surface expression in 1,040 AML patients, while PMID 38468826 compares MFI between groups in a smaller single-centre B-ALL series. **No study has measured CD123 in both lineages under one protocol**, so the inversion is inferred from non-comparable measurements rather than observed directly. What is solid is that the AML result cannot be used to license a CD123 rationale in infant B-ALL.

    Evidence that argues against this (1)
    • Lower MFI is not absence, and antibody-drug conjugates and CAR products can be effective against antigens expressed at low density - so a lower CD123 MFI weakens the enrichment argument without ruling out CD123-directed therapy. The honest reading is that the strategic case must be made on B-lineage data that does not yet exist, not that the target is dead. Bisegna ML et al. 2024
Connected to