Oncofusion

start / graph / concept-immunophenotype-as-a-genotype-surrogate

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

What an immunophenotype can and cannot tell you about a genotype - the arithmetic of a surrogate marker, worked on NG2

also called surrogate marker arithmetic · flow cytometry versus genotype · NG2 predictive value · immunophenotypic prediction of KMT2A rearrangement

A flow cytometer reads proteins on the outside of a cell in a few hours, while a genetic test reads the broken gene inside it in a few days - so people use the proteins to guess the gene. This entry sets out exactly how good that guess is, and the honest answer is that it depends entirely on which direction you are reasoning in and how common the genotype is in the patients you see.

The same thing, for a scientist

This graph has the panel (`concept-flow-cytometry-immunophenotype-kmt2a-r`), the blind spot (`concept-mrd-flow-cytometry-and-the-lineage-switch-blind-spot`) and now the markers. What it has not had is the **arithmetic that connects a marker to a genotype**, and every published number about NG2 in this disease is a different quantity from the one a clinician usually wants. **Four designs, four different questions, four numbers that are all correct.** | design | starting population | what it measures | number | |---|---|---|---| | Behm 1996 | 104 consecutive childhood ALL | screening in an unselected cohort | 9/9 NG2+ had MLL-r; 0/95 NG2- did | | Wuchter 2000 | 533 leukaemias, conditioned on genotype | **sensitivity** within MLL-r | 58/72 = **81%** | | Schwartz 2003 | 313 adult ALL, MLL status in 120 B-precursor | sensitivity, specificity, PPV together | **0.89 / 0.89 / 0.93** | | Bisegna 2024 | 1,382 B-ALL, conditioned on the marker | **positive predictive value** | 85/88 = **96.6%** | **The 81% and the 96.6% are not in conflict and are not comparable.** One conditions on the genotype and asks how often the marker is there; the other conditions on the marker and asks how often the genotype is there. A reader who takes the larger number as "how good NG2 is" has silently changed the question. **Positive predictive value is not a property of a marker.** With sensitivity 0.89 and specificity 0.89 [E2], PPV depends on prevalence: at the roughly 10% KMT2A-r prevalence of a general paediatric ALL population, those figures give a PPV near 0.47 - about a coin flip - while a referral series enriched for the genotype returns 0.93 or 0.97. **Bisegna's 96.6% is a property of a Rome referral centre's case mix.** The same instrument in a general children's hospital gives a materially worse answer. **Three further limits that hold for every marker in this beat.** 1. **A threshold is not a constant.** ROC-optimal NG2 cut-offs range from **7% (infant ALL) to 28% (infant AML)** [E3]. A single published cut-off transported across lineage or age is a different test. 2. **The laboratory is part of the test.** In 148 children under 24 months across multiple Brazilian centres, predictive accuracy **varied greatly per centre** and the authors concluded that staff training and standardised sampling are essential [E4]. 3. **Multi-marker scores do not obviously beat the best single marker.** The ROC-derived five-marker t(11q23) score (NG2, CD10, CD15, CD34, CD20) reaches sensitivity 84.21% and specificity 99.04% in its derivation cohort [E5] - and NG2 alone reached 0.89/0.89 in a different one. A score derived and evaluated in the same cohort is optimistic by construction. **And the limit that matters most in this disease: an immunophenotype is a state, a genotype is not.** `gene-cspg4` is a direct MLL-AF4 target gene, `gene-prom1` is driven by an H3K79me2/3 element, `gene-mme` is silenced by promoter methylation. All three are outputs of a transcriptional programme. **Anything that moves the programme moves the marker while the rearrangement stays exactly where it is** - steroid before the flow sample already does it [E6], and 52% of revumenib-treated patients showed dynamic immunophenotypic change on sequential flow [E7].

Go deeper
The rule this entry proposes for the graph

Any entry quoting a marker's performance must say which of these it means:

  • Sensitivity - of patients with the genotype, how many have the marker. Answers: if I skip the marker, who do I miss?
  • Specificity - of patients without the genotype, how many lack the marker.
  • Positive predictive value - of patients with the marker, how many have the genotype. Prevalence-dependent, therefore centre-dependent. Answers: this patient is marker-positive; should I act?
  • Negative predictive value - almost never reported in this literature and it is what a triage decision actually rests on.

And it must say the population the number came from, because in this beat the same marker's number changes with lineage (B versus myeloid), age (infant versus adult), assay threshold and laboratory.

The forward-looking version. A surrogate marker is worth having when it is fast and the confirmatory test is slow. NG2 by flow returns in hours; FISH and RT-PCR in days. That gap is the marker's entire value, and it is narrowing: rapid targeted sequencing and optical genome mapping shorten the confirmatory arm. A surrogate's worth is a function of the confirmatory test's turnaround, not of the surrogate's biology, and no entry in this graph had said so.

Facts and figures
prevalence notePPV is prevalence-dependent. At sens 0.89 and spec 0.89, a ~10% KMT2A-r prevalence gives PPV near 0.47; a referral series enriched for the genotype gives 0.93-0.97. The 0.47 is arithmetic performed in this entry on the published sensitivity and specificity, not a measured value.
designs compared
8562939104.08562939: 10410914547533.010914547: 533384688261,38238468826: 1382
designs compared — the numbers behind the chart
[
  {
    "pmid": "8562939",
    "n": 104,
    "conditioning": "unselected consecutive childhood ALL",
    "quantity": "screening performance",
    "result": "9/9 NG2+ MLL-r; 0/95 NG2- MLL-r",
    "title": "Human homologue of the rat chondroitin sulfate proteoglycan, NG2, detected by monoclonal antibody 7.1, identifies childhood acute lymphoblastic leukemias with t(4;11)(q21;q23) or t(11;19)(q23;p13) and MLL gene rearrangements.",
    "authors": "Behm FG, Smith FO, Raimondi SC, et al.",
    "venue": "Blood",
    "year": 1996,
    "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."
  },
  {
    "pmid": "10914547",
    "n": 533,
    "conditioning": "on genotype",
    "quantity": "sensitivity",
    "result": "58/72 = 81% at 20% cut-off",
    "title": "Detection of acute leukemia cells with mixed lineage leukemia (MLL) gene rearrangements by flow cytometry using monoclonal antibody 7.1.",
    "authors": "Wuchter C, Harbott J, Schoch C, et al.",
    "venue": "Leukemia",
    "year": 2000,
    "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."
  },
  {
    "pmid": "12886247",
    "n": "313 (120 with MLL status)",
    "conditioning": "adult B-precursor ALL",
    "quantity": "sens/spec/PPV",
    "result": "0.89 / 0.89 / 0.93",
    "title": "Expression of the human homologue of rat NG2 in adult acute lymphoblastic leukemia: close association with MLL rearrangement and a CD10(-)/CD24(-)/CD65s(+)/CD15(+) B-cell phenotype.",
    "authors": "Schwartz S, Rieder H, Schläger B, et al.",
    "venue": "Leukemia",
    "year": 2003,
    "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."
  },
  {
    "pmid": "38468826",
    "n": 1382,
    "conditioning": "on the marker",
    "quantity": "positive predictive value",
    "result": "85/88 = 96.6% at 10% cut-off",
    "title": "NG2 Molecule Expression in Acute Lymphoblastic Leukemia B Cells: A Flow-Cytometric Marker for the Rapid Identification of KMT2A Gene Rearrangements.",
    "authors": "Bisegna ML, Peragine N, Elia L, et al.",
    "venue": "Mediterr J Hematol Infect Dis",
    "year": 2024,
    "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."
  }
]
threshold dependence
{
  "infant_ALL": 7,
  "child_ALL": 12,
  "infant_AML": 28,
  "child_AML": 14,
  "pmid": "33135273",
  "title": "Quantification of NG2-positivity for the precise prediction of KMT2A gene rearrangements in childhood acute leukemia.",
  "authors": "Zerkalenkova E, Mikhaylova E, Lebedeva S, et al.",
  "venue": "Genes Chromosomes Cancer",
  "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."
}
state dependence
[
  "steroid before the diagnostic sample (PMID 38468826, inference on 2 cases)",
  "menin inhibition: 16/31 (52%) with dynamic immunophenotypic change (PMID 41814014)",
  "extramedullary/CNS site: NG2 upregulated (PMID 28943635)"
]
How this could bite us back
  • Triaging fusion testing on a flow surrogate will miss t(4;11). NG2's sensitivity of 0.81-0.89 means one in six to one in nine rearranged patients is marker-negative, and eight MLL-AF4-positive patients were NG2-negative in a single adult series. A protocol that uses a marker to decide whether to run FISH must state its plan for those patients.
  • A published positive predictive value is a property of the centre that published it. Transporting 96.6% from a referral laboratory to a general children's hospital overstates the marker by a wide margin, because PPV falls with prevalence while sensitivity and specificity stay put.
  • Every marker in this beat that has a known mechanism is a transcriptional output of the fusion's programme, so any programme-modifying therapy - steroid, a menin inhibitor, a DOT1L inhibitor, a hypomethylating agent - can move the marker without moving the disease. Fifty-two per cent of revumenib-treated patients with sequential flow showed immunophenotypic change.
  • The value of a fast surrogate is the confirmatory test's turnaround, and that is shrinking. An argument for NG2-based triage that was correct when karyotype took two weeks may not survive rapid targeted sequencing, and no entry in this graph had made the dependency explicit.
Still open
  1. What is the negative predictive value of NG2 for KMT2A rearrangement in infants specifically? It is the number a triage decision rests on and no paper found in this beat reports it.
  2. Does a prospective, standardised multi-centre evaluation of the five-marker t(11q23) flow score beat NG2 alone? Both derivation-set numbers exist; the head-to-head does not.
  3. How much does steroid exposure before the diagnostic marrow shift NG2, quantitatively? It is offered as an explanation for false negatives and has never been measured paired.
  4. At what confirmatory-test turnaround does a flow surrogate stop being worth its false negatives? This is a decision-analytic question nobody in this literature has posed.
Every source behind this entry (7)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Among 1,382 newly diagnosed B-ALL cases at a single institution, 88 were NG2-positive at a >10% threshold and 85 of those 88 (96.6%) harboured KMT2A rearrangement, mainly pro-B ALL (77 of 85, 91%); only 2 KMT2A-rearranged B-ALL cases showed NG2 below 10%, attributed to steroid therapy given before the analysis. In a separate 533-sample series conditioned on genotype, 58 of 72 (81%) MLL-rearranged samples were 7.1-positive at a 20% cut-off.

    PMID 38468826DOI
    What this evidence cannot show

    Abstract read in full from raw efetch XML on 2026-09-20; full text not read. **The 96.6% is a positive predictive value in one Rome referral centre's case mix and is not transferable.** The design conditions on NG2 positivity and therefore cannot produce a sensitivity. The 81% from the second series is quoted here from a different paper (PMID 10914547) whose own record was fetched separately on the same day; the two are placed side by side deliberately to make the conditioning difference visible.

    Evidence that argues against this (1)
    • The prevalence dependence is the whole point: with sensitivity and specificity both 0.89 and a KMT2A-r prevalence around 10% in general paediatric ALL, PPV falls to roughly 0.47. A published PPV without its cohort's prevalence is uninterpretable, and this is the commonest way marker performance is misquoted. Schwartz S et al. 2003
  2. E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    In 313 adult ALL patients with MLL rearrangement determined in 120 B-precursor cases by RT-PCR and FISH, NG2 showed sensitivity 0.89, specificity 0.89 and positive predictive value 0.93 for MLL rearrangement. NG2 was positive in three patients without detectable MLL rearrangement and negative in eight patients with MLL-AF4 transcripts, but 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.** The PPV of 0.93 is itself prevalence-dependent and reflects this cohort's MLL-r frequency among B-precursor ALL, which is higher than in unselected childhood ALL. The 'prevalence around 10%' figure used in the prevalence recomputation in this entry is a general characterisation of childhood ALL and is not measured in any paper cited here - the recomputation is arithmetic on the reported sensitivity and specificity and is flagged as such.

    Evidence that argues against this (1)
    • The 100%-accuracy subgroup claim (CD65s+ and/or CD15+) is the same optimism in another form: a subgroup selected after the fact in which the marker performed perfectly is a description of this cohort, not a validated rule, and no independent replication of that subgroup rule was found in this beat. Emerenciano M et al. 2011
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    ROC analysis on 505 paediatric acute leukaemia patients with NG2 positivity above 1% established threshold levels for NG2 positivity of 7% in infants with ALL, 12% in older children with ALL, 28% in infants with AML and 14% in patients over 1 year with AML, and these thresholds improved diagnostic performance over the conventional 10% and 20% cut-offs.

    PMID 33135273DOI
    What this evidence cannot show

    Abstract read in full from raw efetch XML on 2026-09-20; full text not read. **Thresholds derived and evaluated in the same cohort with no stated independent validation set**, so the improvement over conventional cut-offs is optimistic by construction. The cohort is conditioned on NG2 above 1%, which is not the population a threshold would be applied to. Single-country referral series.

    Evidence that argues against this (1)
    • Four different optimal thresholds for one marker is itself an argument that the marker is reading something other than the genotype - a threshold that depends on lineage and age is a threshold on a state variable. This is the observation that motivates hyp-ng2-tracks-the-transcriptional-programme-not-the-rearrangement. Lopez-Millan B et al. 2024
  4. E4established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    In 148 patients under 24 months of age (86 under 12 months) from various centres in Brazil, NG2 could be used for MLL rearrangement prediction, but the accuracy of the prediction varied greatly on a per-centre basis, and the authors concluded that proper staff training and standardised sampling procedures are essential when receiving samples from multiple centres.

    PMID 21444110DOI
    What this evidence cannot show

    Abstract read in full from raw efetch XML on 2026-09-20; full text not read. **The abstract gives no per-centre numbers**, so 'varied greatly' is the authors' characterisation and this entry cannot quantify the spread. n = 148 across an unstated number of centres. This is the population this graph cares about most - children under 24 months, 86 under 12 months - which is why it is carried despite the missing numbers.

    Evidence that argues against this (1)
    • A multi-centre accuracy problem could be a reagent, gating or sample-handling problem rather than a property of the marker, in which case standardisation fixes it and the marker is fine. The paper's own conclusion points that way. The uncomfortable version is that standardisation has been recommended since 2011 and the 2024 single-centre PPV of 96.6% is still a single-centre number. Bisegna ML et al. 2024
  5. E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    In 377 retrospectively analysed B-ALL patients with cytogenetics by interphase FISH, ROC analysis produced flow cytometric predictive scoring systems using NG2, CD10, CD15, CD34 and CD20 for t(11q23) with sensitivity 84.21% and specificity 99.04%; CD27, CD9, CD66c, CD10, CD25 and CD34 for t(12;21) with 89.29% and 96.15%; CD25, CD38, CD34 and CD66c for t(9;22) with 75.00% and 88.19%; and CD34, cytoplasmic mu, CD123 and CD66c for t(1;19) with 85.71% and 92.71%. The authors call for prospective evaluation.

    PMID 30730614DOI
    What this evidence cannot show

    Abstract read in full from raw efetch XML on 2026-09-20; full text not read. **Retrospective, single-centre, n = 377, and all four scores were derived and evaluated in the same cohort** - the authors themselves ask for prospective evaluation, so every number here is a derivation-set number. The t(9;22) score at 75.00% sensitivity is the reminder that not every fusion has an NG2: a flow score is only as good as whether the genotype has a surface consequence.

    Evidence that argues against this (1)
    • The t(11q23) score's 84.21% sensitivity does not clearly beat NG2 alone at 0.89 in a different cohort. Adding four markers to the best single marker and getting no obvious gain is a warning that the extra markers are carrying derivation-set noise, and only a prospective head-to-head would settle it. Schwartz S et al. 2003
  6. E6strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    NG2 is an epigenetically regulated direct target gene of the MLL-AF4 fusion protein, negatively regulates the glucocorticoid receptor NR3C1 through FLT3 and AP-1-mediated transrepression, and confers glucocorticoid resistance to MLLr B-ALL cells. In a separate series, the only two KMT2A-rearranged B-ALL cases with NG2 below the 10% threshold were attributed by the authors to steroid therapy administered before the cytofluorimetric analysis.

    PMID 39093982DOI
    What this evidence cannot show

    Abstract read in full from raw efetch XML on 2026-09-20; full text not read. The steroid attribution in the second sentence comes from a different paper (PMID 38468826, fetched the same day) and is **the authors' inference about two patients, not a measurement** - no study found in this beat has measured NG2 before and after steroid in the same patients. The mechanistic half is cell-line work.

    Evidence that argues against this (1)
    • If steroid exposure lowers NG2 below threshold, then the marker's reported sensitivity is itself a function of when the sample was drawn relative to the first dose - a pre-analytical variable that none of the four performance studies in this entry reports or controls for. Bisegna ML et al. 2024
  7. E7strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    In sequential bone marrow specimens from 48 patients with relapsed or refractory AML treated with the menin inhibitor revumenib, dynamic changes in the immunophenotype after treatment were observed in 16 of 31 (52%) patients, characterised by a switch from a myeloid/stem-like to a monocytic or myelomonocytic immunophenotype or vice versa, or by substantial changes in the intensity of antigen expression or in patterns of leukaemia-associated immunophenotypes. Morphologic remission with undetectable flow MRD was associated with median overall survival of 23.6 months, compared with 20.8 months with detectable MRD and 3.2 months in non-responders. The authors conclude that flow-cytometric treatment monitoring after menin inhibition requires recognition of phenotypic changes associated with differentiation.

    PMID 41814014DOI
    What this evidence cannot show

    Abstract read in full from raw efetch XML on 2026-09-20; full text not read; PMCID PMC13233299 exists so the five-route test remains available to a reader wanting the marker-by-marker detail. **Note the denominators differ within the abstract: 48 patients treated, 31 with sequential specimens evaluable, 16 changed.** The 52% is of 31, not of 48. **Adult relapsed/refractory AML**, not infant B-ALL, and the abstract does not separate KMT2A-rearranged from NPM1-mutant patients. The survival figures compare non-randomised response groups.

    Evidence that argues against this (1)
    • These changes are attributed to differentiation, which is the drug's intended mechanism - so a phenotype shift under menin inhibition is a response signature rather than evidence that markers are untrustworthy. The problem it creates for surrogate markers is real either way: a panel calibrated at diagnosis is not calibrated on treatment. Kovach AE et al. 2026
Connected to