Oncofusion

start / graph / gene-cd33

Genestrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.Q45-immunophenotype

CD33 (SIGLEC3) - the aberrant myeloid antigen on a B-lineage blast, and the one antibody-drug-conjugate target with a positive randomised result in KMT2A-rearranged disease

also called CD33 · SIGLEC3 · Siglec-3 · Myeloid cell surface antigen CD33 · gp67 · P20138

CD33 is a myeloid marker that KMT2A-rearranged leukaemia cells often wear even when they are otherwise B-lineage, which is one of the oddities that made people suspect the disease is lineage-confused from the start. It is also a drug target: adding a CD33-directed antibody-drug conjugate to chemotherapy improved event-free survival specifically in children with KMT2A-rearranged AML.

The same thing, for a scientist

CD33 (UniProt **P20138**, CD33_HUMAN, **364 aa**, single reviewed human record, synonym SIGLEC3) is a sialic-acid-binding Ig-like lectin on myeloid progenitors and monocytes. In this graph it is named in 30 entries and has never had one. The only entry that covers it, `target-cd33-cd123-myeloid-directed-kmt2a-r`, **covers two different targets in one entry**, which the PI's standing granularity instruction forbids; this entry and `gene-il3ra` split it. **Two distinct roles, and they should not be conflated.** *As a diagnostic oddity.* KMT2A-rearranged paediatric ALL characteristically co-expresses the myeloid antigens CD15, CD33, CD65 and CD68 on a CD19-positive, usually CD10-negative blast [E1]. This aberrant co-expression is part of the EGIL BI pro-B pattern that raises the KMT2A hypothesis at the flow bench. **But CD33 is the weakest member of that set for this purpose:** in adult B-precursor ALL, NG2 positivity - the marker that actually predicts the rearrangement - was associated with a **CD13-negative, CD33-negative** phenotype [E2], and in a 105-child B-ALL series CD33 positivity did not differ between CD34-positive and CD34-negative groups [E3]. CD15 and CD65 carry the signal; CD33 rides along. *As a therapeutic target with real randomised evidence.* In COG AAML0531, **215 of 1,022 (21%) enrolled patients had KMT2A-rearranged AML**. Adding gemtuzumab ozogamicin gave **EFS 48% vs 29% (P = .003)**, lower relapse risk among complete responders (**40% vs 66%, P = .001**), 5-year DFS **57% vs 33% (P = .002)**, and in transplanted patients 5-year relapse risk **28% vs 73% (P = .006)**; OS did not reach significance (63% vs 53%, P = .054) [E4]. That is the strongest randomised signal for any antigen-directed agent in KMT2A-rearranged disease in this graph. **The bridge between the two roles is the thing to be careful about.** AAML0531 is **AML**. The CD33 co-expression that makes KMT2A-rearranged *B-ALL* interesting is a different clinical situation, and no randomised trial has tested a CD33-directed agent in infant KMT2A-rearranged B-ALL.

Go deeper
Why an aberrant myeloid antigen on a B blast matters to this graph

CD15/CD33/CD65 co-expression on a CD19-positive pro-B blast is not a laboratory curiosity - it is the immunophenotypic shadow of the lineage plasticity that this graph's mechanism-lineage-switch-kmt2a and hyp-lineage-switch-is-competitive-release-of-preexisting-clone are about. The blast is already displaying the myeloid programme at diagnosis, before any antigen-directed pressure is applied.

Two readings, and the literature supports the weaker one. The strong reading is that myeloid-antigen-positive KMT2A-r B-ALL is primed to switch and the aberrant antigens predict it. The weak reading is that co-expression is common in KMT2A-rearranged disease and in B-ALL generally, and is not by itself predictive. The iBFM CD371 study is the closest clean test of the strong reading: CD371 - a different myelomonocytic marker - identified 160 of 1,812 BCP-ALL (8.8%), and those cases showed a transient myelomonocytic switch in up to 65.4% of day-15 samples and slow early response, yet 5-year EFS was 88.3% 5. A marker that flags a real, measurable, transient myeloid shift, and does not flag a bad outcome. CD33 is not CD371 and this is not the same claim, but it is the right calibration: aberrant myeloid antigen expression in BCP-ALL is common, sometimes dynamic, and mostly not prognostic.

Cell type, cell state, localisation, epigenetics
  • Cell type. Myeloid progenitors, monocytes, and a large fraction of normal CD34-positive haematopoietic stem and progenitor cells - which is the whole toxicity story. Aberrantly on KMT2A-rearranged B-lineage blasts 1.
  • Cell state. Present at diagnosis on the B-lineage blast, so it is not a switch-acquired antigen; it is a persistent lineage-infidelity feature.
  • Localisation. Plasma membrane, type I transmembrane, 364 residues 6. Rapidly internalised on antibody binding, which is why it works as an ADC target rather than a naked-antibody target.
  • Epigenetics. Nothing in this beat establishes CD33 as a direct KMT2A-fusion target gene, unlike gene-cspg4 and gene-prom1. The honest statement is that CD33 expression tracks the myeloid-primed state the fusion produces.
Facts and figures
uniprotP20138
gene symbolCD33
length AA364
cd numberCD33
familySIGLEC (SIGLEC3); note SIGLEC2 is CD22, also in this beat - a rule-63 paralogue-collision hazard
accession resolutiongene_exact:CD33 AND organism_id:9606 AND reviewed:true, fields include gene_primary and length; one row; primary name read back as CD33; length 364 asserted for verify-accessions.py.
subnuclear localizationNOT APPLICABLE. CD33/Siglec-3 is a single-pass type I transmembrane cell-surface receptor (signal peptide 1-17, transmembrane 260-282) with no nuclear pool in UniProt, no annotated NLS, and no reported nuclear localisation in any system. Its alternative compartment is the PEROXISOME, not the nucleus. Recorded NOT APPLICABLE so it is excluded from the denominator of unmeasured subnuclear questions, as Q87 did for B2M and PROM1 in tranche 2. The dispatch brief predicted this outcome for CD33 and the prediction is confirmed - with the addition that 'cell-surface receptor' undersells the record, because the second isoform is intracellular.
localization shown for fusionPARTIAL, and the partial is precise. CD33 SURFACE EXPRESSION is documented on KMT2A-rearranged blasts in both lineages (this entry's E1 and E4). The CD33M-surface-versus-CD33m-peroxisome ISOFORM SPLIT has never been measured in a KMT2A-rearranged blast specifically: Siddiqui et al. worked in normal blood neutrophils, monocytes and a microglial line [E_q89_cd33_loc2], and the AAML0531 genotype analysis reports surface CD33 intensity across all de novo paediatric AML without a KMT2A-rearranged stratum in its abstract, although 215 of that trial's 1,022 patients (21%) were KMT2A-rearranged by this entry's existing E4 [E_q89_cd33_loc3].
randomised evidence
{
  "trial": "COG AAML0531",
  "population": "215/1022 (21%) paediatric KMT2A-rearranged AML",
  "efs_with_go_pct": 48,
  "efs_without_go_pct": 29,
  "efs_p": 0.003,
  "os_with_go_pct": 63,
  "os_without_go_pct": 53,
  "os_p": 0.054,
  "relapse_risk_cr_with_go_pct": 40,
  "relapse_risk_cr_without_go_pct": 66,
  "pmid": "34048275",
  "design_caveat": "subgroup of a randomised trial, not a randomised trial in this subgroup",
  "title": "Gemtuzumab Ozogamicin Improves Event-Free Survival and Reduces Relapse in Pediatric KMT2A-Rearranged AML: Results From the Phase III Children's Oncology Group Trial AAML0531.",
  "authors": "Pollard JA, Guest E, Alonzo TA, et al.",
  "venue": "J Clin Oncol",
  "year": 2021,
  "title_provenance": "Title/authors/venue backfilled from PubMed esummary on 2026-09-20 because this citation stored a bare PMID with no title, which made it unscoreable by verify-citations.py. BACKFILLING IS NOT VERIFICATION: it confirms the PMID resolves to a real paper, not that this is the right paper for the claim. That still requires reading."
}
depmap kmt2a selectivity
{
  "chronos_mean_kmt2a_rearranged": 0.1162,
  "chronos_mean_other_leukaemia": 0.0283,
  "difference": 0.0879,
  "pan_essentiality_all_lines": 0.0585,
  "selectivity_rank": 16600,
  "rank_denominator": 17728,
  "provenance": "Joined 2026-09-21 from analyses/depmap-kmt2a-selective-scan-full (generated 2026-09-19T10:12:06Z), the project's own genome-wide scan over 11 KMT2A-rearranged and 43 other leukaemia DepMap lines. NOT re-fetched and NOT independent of that scan: every caveat it carries applies here, including that the KMT2A-rearranged arm is a small number of lines drawn from nine individuals, that the rank denominator is 17,728 genes, and that a difference is not a dependency. Added because only 26 of 308 gene entries carried this number, which made any KMT2A-selectivity question answerable for 8% of the corpus.",
  "how_to_read": "A negative difference means MORE required in KMT2A-rearranged lines than in other leukaemia lines. A rank near 17,728 is ANTI-selective. Neither is evidence of a fusion-driven dependency on its own: check pan_essentiality_all_lines before calling anything selective, and note that a paralogue-redundant gene can score near zero for reasons that have nothing to do with this disease - see ana-the-paralogue-redundancy-excuse-is-not-tested-and-my-proxy-decided-the-answer."
}
localization
{
  "subcellular": "PLASMA MEMBRANE - but ISOFORM-SCOPED, and the second isoform is not there. UniProt P20138 fetched 2026-09-21 carries TWO isoform-scoped SUBCELLULAR LOCATION comments. Isoform CD33M: 'Cell membrane', single-pass type I membrane protein, ECO:0000269 PubMed:10611343 and PubMed:28747436, with a signal peptide at 1-17 and a helical transmembrane segment at 260-282. Isoform CD33m: 'PEROXISOME', ECO:0000269 PubMed:28747436, with the uncoded note that 'CD33m isoform does not localize to cell surfaces but instead accumulates in peroxisomes' [E_q89_cd33_loc1].",
  "why_this_is_not_a_footnote": "The exon-2-skipped isoform D2-CD33/CD33m lacks the IgV sialic-acid-binding domain, which is the gemtuzumab-ozogamicin epitope AND the epitope most diagnostic flow panels score. So the single most important fact about CD33's localisation in this disease is that a common germline splice variant moves the protein OFF the surface and into a peroxisome, where neither the drug nor the flow cytometer can see it [E_q89_cd33_loc2].",
  "reading_note_rule_91": "This is a case where reading the UNCODED and ISOFORM-SCOPED comments mattered. Filtering P20138 on ECO:0000269 alone still returns both rows here, but filtering on the FIRST location value - the reflex - returns 'Cell membrane' and silently discards the peroxisomal isoform. The same shape as KMT2A, where the N320 and C180 fragment comments are the interesting ones and are uncoded."
}
cell type
{
  "normal": "MYELOID LINEAGE. UniProt records the tissue specificity as 'Monocytic/myeloid lineage cells. In the brain, CD33 is mainly expressed on microglial cells'. Siddiqui et al. measured the isoform split in primary human BLOOD NEUTROPHILS and MONOCYTES and in a microglial cell line [E_q89_cd33_loc1][E_q89_cd33_loc2].",
  "in_this_disease": "ABERRANTLY ON THE CD19-POSITIVE KMT2A-REARRANGED PRO-B BLAST, and on KMT2A-rearranged AML blasts. This is the entry's existing E1 finding and the reason the gene is here; the field records it so the coverage reporter can see it.",
  "the_toxicity_cell_type": "Normal CD34-positive haematopoietic stem and progenitor cells, which is why a CD33-directed ADC is myelosuppressive - already stated in this entry's prose and repeated here because the field is what gets machine-read."
}
cell state
{
  "state": "PRESENT AT DIAGNOSIS on the B-lineage blast, so it is a persistent lineage-infidelity feature rather than a switch-acquired antigen; and see cell-state-differentiation-block, the state the aberrant myeloid antigen is a marker of.",
  "a_genotype_is_not_a_cell_state": "The rs12459419 CC/CT/TT split changes how much CD33 reaches the surface [E_q89_cd33_loc2], but it is a germline genotype, not a cell state - the blast does not switch between them. Recorded here so a later reader does not file it as one."
}
localization bounded absence
{
  "date": "2026-09-21",
  "tool": "NCBI esearch db=pubmed",
  "queries": [
    {
      "query": "CD33[tiab] AND rs12459419[tiab]",
      "querytranslation": "\"CD33\"[Title/Abstract] AND \"rs12459419\"[Title/Abstract]",
      "count": 15
    },
    {
      "query": "CD33[tiab] AND (KMT2A[tiab] OR MLL[tiab])",
      "querytranslation": "\"CD33\"[Title/Abstract] AND (\"KMT2A\"[Title/Abstract] OR \"MLL\"[Title/Abstract])",
      "count": 75
    }
  ],
  "reading": "Fifteen records on the splicing polymorphism and seventy-five on CD33 in KMT2A/MLL disease, and this beat found no record joining the two - i.e. no published analysis of the CD33 splice genotype or the surface-versus-peroxisomal isoform ratio within a KMT2A-rearranged cohort. Titles were screened; full texts were not, so this is bounded by titles and abstracts (rule 80) and a stratified analysis could exist inside a supplementary table.",
  "absence_class": "measured-not-reported"
}
How this could bite us back
  • CD33 is on normal CD34-positive haematopoietic stem and progenitor cells, so a CD33-directed ADC is a myelosuppressive agent by design. In AAML0531 prior GO exposure preceded transplant in a large fraction of the benefiting patients, and the population this graph cares about is infants, whose marrow reserve and whose subsequent transplant conditioning are both less forgiving than an older child's.
  • Gemtuzumab ozogamicin carries hepatic sinusoidal obstruction syndrome risk, and the KMT2A-r subgroup analysis reports efficacy without a subgroup-specific toxicity denominator. The sibling ADC in this beat, inotuzumab, produced grade 3 SOS in 6 of 21 children (28.6%) proceeding to transplant, so the class hazard is real and is worst exactly where the KMT2A-r benefit was largest - around transplant.
  • Targeting CD33 in KMT2A-rearranged B-ALL would be targeting an antigen the blast carries as a minority programme. If the myeloid programme is the escape route rather than the disease, a CD33-directed agent might pre-emptively suppress the switch - or might select for the CD33-negative B-lineage clone and leave the patient exactly where they started. Neither has been tested.
  • The randomised evidence is a subgroup analysis with a non-significant overall survival (P = .054). Any decision that treats the CD33 result as established should state which endpoint it is relying on.
Still open
  1. Does CD33 co-expression at diagnosis predict lineage switch in KMT2A-rearranged infant B-ALL? The CD371 study shows how to answer this - a large prospectively immunophenotyped cohort with long follow-up - and nobody has run it for CD33 in this genotype.
  2. Is the adult CD33-negative association within NG2-positive ALL an age effect or a cohort artefact? It contradicts the paediatric convention and a single well-designed paediatric measurement would settle it.
  3. Would a CD33-directed agent given at diagnosis in infant KMT2A-r B-ALL prevent or accelerate the myeloid switch? This is a testable question in xenograft and the graph has no entry proposing it.
  4. What is the CD33 splice-variant distribution in KMT2A-rearranged infants, and does it match the AML populations in which gemtuzumab was tested?
  5. Is the rs12459419 genotype distribution, or the D2-CD33 transcript fraction, different in KMT2A-rearranged AML than in other paediatric AML? AAML0531 genotyped 816 patients and 21% of that trial was KMT2A-rearranged, so the stratified analysis is a table away and would say whether the one positive randomised ADC result in this disease is genotype-restricted.
  6. Has anyone measured the CD33M-surface versus CD33m-peroxisome ratio in a KMT2A-rearranged blast, in either lineage? The isoform split was measured only in normal neutrophils, monocytes and microglia.
  7. Does the aberrant CD33 on a CD19-positive KMT2A-rearranged pro-B blast use the same isoform balance as CD33 on a myeloid blast, or does the lineage-infidelity programme reach the splice enhancer too?
Every source behind this entry (9)
  1. E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.review

    KMT2A-rearranged paediatric ALL typically shows an immature pro-B immunophenotype with expression of CD19 and CD34 and co-expression of CD15, CD33, CD65 and CD68, often lacking CD10, and expresses NG2/CSPG4 in about 90% of cases.

    PMID 36979800DOI
    What this evidence cannot show

    **This is a narrative review, cited as a map rather than as a measurement.** Abstract read in full from raw efetch XML on 2026-09-20; full text not read. The '90% NG2' figure in this review is higher than the primary measurements (81% sensitivity at n=533, 0.89 in adult B-precursor ALL) and this entry does not rely on it. Confidence is moderate because review-level marker lists smooth over the age, lineage and cohort dependence that the primary papers make explicit.

    Evidence that argues against this (1)
    • Myeloid-antigen co-expression is frequent across childhood ALL generally and not specific to KMT2A rearrangement: in 283 de novo childhood ALL cases, at least one of CD13, CD33 or CD15 was expressed in 64% of B-lineage and 74% of T-lineage cases. The patterns are nonrandom by genotype and the MLL-AF4-specific finding in that series is the ABSENCE of CD13, not the presence of CD33. A marker list that bundles CD13 with CD33 for this genotype is therefore wrong in at least one member. Abdelhaleem M 2007
  2. E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    In adult B-precursor ALL, NG2 expression was significantly associated with a CD10-negative, CD34-negative, CD24-negative, CD65s-positive, CD15-positive, CD13-negative, CD33-negative phenotype, and NG2 predicted MLL rearrangement with 100% accuracy among patients with a CD65s-positive and/or CD15-positive immunophenotype.

    PMID 12886247DOI
    What this evidence cannot show

    Abstract read in full from raw efetch XML on 2026-09-20; full text not read. **Adult ALL**, and the CD33-negative association may not hold in infants, where the review-level convention is the opposite [E1]. The association is at the level of the NG2-positive group, not a per-patient rule. Cited here specifically because it separates CD15/CD65 from CD13/CD33 within the same cohort and the same analysis, which a marker list cannot do.

    Evidence that argues against this (1)
    • The standard paediatric description of this disease lists CD33 alongside CD15 and CD65 as co-expressed myeloid antigens. This entry does not resolve the disagreement and states both; if the difference is real it is an age effect and is itself worth measuring. Gorecki M et al. 2023
  3. E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    In 105 children with newly diagnosed B-ALL divided by CD34 positivity at a 10% threshold, no significant differences were found in the positive rates of CD10, CD19, CD20, CD22, CD79a, CD13, CD33 or CD38 between the CD34-positive and CD34-negative groups, although the CD34-negative group was enriched for early pre-B-ALL, E2A-PBX1 and MLL-AF4.

    PMID 38766795DOI
    What this evidence cannot show

    Abstract read in full from raw efetch XML on 2026-09-20; full text not read. **Single-centre, n = 105, CD34-negative arm of only 18**, and a journal of limited standing. A null result on CD33 in 18 patients has almost no power, so this is cited as an absence of a signal in a small series and not as evidence of no difference. The abstract's own text is internally odd - it reports CD13 both as significantly higher in the CD34-positive group and as not significantly different - which is a reason to treat the marker comparisons here as weak.

    Evidence that argues against this (1)
    • The abstract reports CD13 in two apparently contradictory places, so the marker-comparison section of this paper is not reliable and the CD33 null should not be leaned on. It is carried only because it is a paediatric dataset that names CD33 in the index genotype's neighbourhood. Zha W et al. 2024
  4. E4established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    Of 1,022 patients enrolled on Children's Oncology Group trial AAML0531, 215 (21%) had KMT2A-rearranged AML, with 5-year EFS 38% and OS 58% from study entry. EFS was superior with gemtuzumab ozogamicin (48% with GO versus 29% without, P = .003) although OS was comparable (63% versus 53%, P = .054). Among patients achieving complete remission, GO was associated with lower relapse risk (40% versus 66%, P = .001) and improved 5-year disease-free survival (57% versus 33%, P = .002). For patients undergoing transplant, prior GO exposure was associated with 5-year relapse risk of 28% versus 73% (P = .006). GO benefit was observed in both higher-risk and not-high-risk KMT2A-r partner subsets, and in multivariable analysis GO was independently associated with improved EFS, improved DFS and reduced relapse risk.

    PMID 34048275DOI
    What this evidence cannot show

    Abstract read in full from raw efetch XML on 2026-09-20; full text not read. **This is a subgroup analysis of a randomised trial, not a randomised trial in KMT2A-r AML** - the randomisation was in the whole AAML0531 population and the KMT2A-r stratum was identified afterwards, so the P values are not protected by the trial's primary design. **Overall survival did not reach significance (P = .054)**, which is the number most likely to be dropped when this result is quoted. The transplant comparison is not randomised. This is **AML, not B-ALL**, and the population is children, not specifically infants.

    Evidence that argues against this (1)
    • A CD33-directed ADC benefit in KMT2A-rearranged AML says nothing about KMT2A-rearranged infant B-ALL, where CD33 is an aberrant antigen on a minority-programme and the competing escape is lineage switch to exactly the myeloid phenotype a CD33 agent would then be treating. Project EVOLVE collected 75 lineage switches after CD19, CD22, CD38 and CD7 targeting, none after CD33. Silbert SK et al. 2025
  5. E5established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    Among 1,812 children with newly diagnosed BCP-ALL enrolled in AIEOP-BFM ALL 2009, 160 (8.8%) were positive for the myelomonocytic marker CD371. CD371-positivity correlated with older age, lower ETV6::RUNX1 frequency, immunophenotypic immaturity and strong expression of CD34 and CD45. During induction these cases showed a transient myelomonocytic switch in up to 65.4% of samples at day 15 and an inferior early response to chemotherapy, yet 5-year event-free survival was 88.3%. CD371 was the most sensitive marker of transient myelomonocytic switch, whose recognition the authors describe as essential for proper flow MRD assessment.

    PMID 38215390DOI
    What this evidence cannot show

    Abstract read in full from raw efetch XML on 2026-09-20; full text not read. **CD371 (CLEC12A) is not CD33**, and this evidence item is cited as a calibration for how to read aberrant myeloid antigens in BCP-ALL, not as a CD33 result. The cohort is BCP-ALL at large; the abstract does not report the KMT2A-rearranged fraction, and infants are largely absent from AIEOP-BFM ALL 2009 by protocol. 'Transient myelomonocytic switch' as measured here is an induction-phase phenotypic shift, not the lineage switch of Project EVOLVE.

    Evidence that argues against this (1)
    • The 88.3% 5-year EFS is the point against reading aberrant myeloid antigen expression as a danger signal: a marker can flag a real, large, reproducible myeloid phenotypic shift and carry essentially no prognostic weight once MRD-directed therapy is applied. Buldini B et al. 2024
  6. E6established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    CD33 is UniProt P20138 (CD33_HUMAN), Myeloid cell surface antigen CD33 / Sialic acid-binding Ig-like lectin 3, 364 amino acids, a single reviewed human record with primary gene name CD33 and gene synonym SIGLEC3.

    What this evidence cannot show

    Fetched 2026-09-20 with `gene_exact:CD33 AND organism_id:9606 AND reviewed:true`, requesting `gene_primary` and `length` per rules 62 and 65; primary name read back; exactly one reviewed row. **CD33 belongs to the SIGLEC family, which is precisely the paralogue-collision shape rule 63 warns about** - SIGLEC3 is CD33 and SIGLEC2 is CD22, so a lookup that matched synonyms loosely could hand back the other target in this same beat. The length 364 is asserted here so `verify-accessions.py` can catch that swap.

    Evidence that argues against this (1)
    • The canonical 364-residue record is not the only CD33 a patient displays. A common SIGLEC3 splice variant lacking the V-set Ig domain removes the epitope that gemtuzumab binds, and CD33 genotype has been reported to modify response to CD33-directed ADCs - so an accession-level identity claim carries no information about whether a given patient's CD33 is targetable. Pollard JA et al. 2021
  7. E_q89_cd33_loc1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    UniProt P20138 (CD33_HUMAN, 364 aa, primary gene CD33, synonym SIGLEC3, read back 2026-09-21) carries TWO ISOFORM-SCOPED subcellular locations: isoform CD33M at the 'Cell membrane' as a single-pass type I membrane protein (ECO:0000269 PubMed:10611343, PubMed:28747436), and isoform CD33m in the 'Peroxisome' (ECO:0000269 PubMed:28747436), with the note that CD33m does not localise to cell surfaces. Signal peptide 1-17; transmembrane helix 260-282. Tissue specificity: monocytic/myeloid lineage cells, and microglia in brain. There is no nuclear location value.

    What this evidence cannot show

    Curated database record; the isoform-scoped rows were read deliberately rather than taking the first location value, because taking the first would have returned 'Cell membrane' alone and lost the peroxisomal isoform entirely. Both rows trace to a single experimental paper for the peroxisomal call, so the peroxisome result currently has one source.

    Evidence that argues against this (1)
    • Reading the UniProt isoform rows as though they were two populations of protein in one cell would be wrong, and the entry should not let that stand. The rows describe the destination of two SPLICE PRODUCTS, and which one a given patient's blasts make is set largely by a germline SNP, so in a CC-homozygous patient the peroxisomal row describes almost nothing and in a TT patient it describes much of the protein. A localisation field that reads 'cell membrane and peroxisome' without the genotype attached is therefore true of the gene and false of most individual cells. That is why this entry carries the genotype in the same breath as the compartment. Lamba JK et al. 2017
  8. E_q89_cd33_loc2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    In primary human blood NEUTROPHILS and MONOCYTES and in a microglial cell line, the exon-2-skipped CD33m isoform DOES NOT localise to cell surfaces and instead accumulates in PEROXISOMES; cell stimulation and activation did not mobilise CD33m to the surface. CD33 is the only Siglec carrying a peroxisome-targeting sequence, a motif that arose by convergent evolution in toothed whales.

    The Alzheimer's disease-protective CD33 splice variant mediates adaptive loss of function via diversion to an intracellular pool. · Siddiqui SS, Springer SA, Verhagen A, Sundaramurthy V, Alisson-Silva F, Jiang W, Ghosh P, Varki A · J Biol Chem 2017
    PMID 28747436DOI
    What this evidence cannot show

    Human primary myeloid cells and a microglial line - the right lineage for this graph but NOT a leukaemic blast and NOT a KMT2A-rearranged cell. The paper's framing is Alzheimer's disease risk, not leukaemia, so it is reachable only by searching the compartment rather than the disease (rule 77 in reverse). Abstract read; the imaging and the CD33m-specific monoclonal antibody were not examined in full text.

    Evidence that argues against this (1)
    • A peroxisomal destination is a surprising claim for a Siglec and deserves an active objection rather than a restatement. Peroxisomal import handles folded proteins through PEX5/PEX7 and does not obviously accommodate a type I membrane protein whose mature form is glycosylated in the secretory pathway, so 'accumulates in peroxisomes' could describe an overexpression artefact or an antibody cross-reaction with a peroxisomal protein rather than genuine targeting. Three things argue against dismissing it: the authors raised a CD33m-SPECIFIC monoclonal antibody instead of relying on a tag, they worked in primary blood cells rather than transfectants, and the evolutionary argument (a peroxisome-targeting motif present in CD33 alone among Siglecs and convergently in toothed whales) is a prediction the artefact hypothesis does not make. It remains a single-laboratory result, and this graph has no independent replication of it. UniProt Consortium 2026
  9. E_q89_cd33_loc3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    In the randomised phase III Children's Oncology Group trial AAML0531 (816 genotyped de novo paediatric AML patients), the rs12459419 T allele was significantly associated with higher D2-CD33 transcript (P < 1.0E-6) and with LOWER DIAGNOSTIC LEUKAEMIC CELL SURFACE CD33 INTENSITY (P < 1.0E-6). Patients with the CC genotype had lower relapse risk on the gemtuzumab-ozogamicin arm than on the no-GO arm (26% v 49%, P < .001) and higher disease-free survival (65% v 46%, P = .004), while CT/TT patients derived no benefit from GO (relapse 39% v 40%, P = .85).

    CD33 Splicing Polymorphism Determines Gemtuzumab Ozogamicin Response in De Novo Acute Myeloid Leukemia: Report From Randomized Phase III Children's Oncology Group Trial AAML0531. · Lamba JK, Chauhan L, Shin M, Loken MR, Pollard JA, Wang YC, Ries RE, Aplenc R, Hirsch BA, Raimondi SC, Walter RB, Bernstein ID, Gamis AS, Alonzo TA, Meshinchi S · J Clin Oncol 2017
    PMID 28644774DOI
    What this evidence cannot show

    Randomised trial, genotype-stratified secondary analysis, so the genotype comparison is not itself randomised. Paediatric de novo AML of all cytogenetics; the abstract does NOT report a KMT2A-rearranged stratum, and this entry's existing E4 records that 215 of 1,022 AAML0531 patients (21%) were KMT2A-rearranged, so the subgroup exists in the cohort but its CD33-genotype analysis is not in this abstract. Abstract read; full text not read. Genotype frequencies are ancestry-dependent and the abstract does not give the ancestry breakdown.

    Evidence that argues against this (1)
    • The clean reading - genotype changes surface epitope density, therefore genotype changes ADC efficacy - is not the only one, and the trial cannot separate the alternatives. rs12459419 sits in a splice enhancer and the same variant has been associated with altered microglial activity and Alzheimer's risk, so the T allele changes CD33 BIOLOGY and not only CD33 EPITOPE AVAILABILITY; a signalling difference in the blast or in the surrounding myeloid compartment could contribute to the differential relapse risk independently of how much drug binds. More sharply for this graph: the surface-intensity measurement is made with a diagnostic flow antibody that, like gemtuzumab, targets the IgV domain, so the assay and the drug are blind in the same way, and 'less antigen' and 'less detectable antigen' cannot be told apart by it (the clone-is-the-identifier problem, rule 74). Nothing in the abstract measures total CD33 protein by a non-IgV reagent. Siddiqui SS et al. 2017
Connected to