NUP98 rearrangements are 7.2% of paediatric AML and 1.6% of adult, with 35% five-year overall survival — and the partner changes everything
also called NUP98-r paediatric AML frequency · NUP98 fusion outcome COG · NUP98 landscape Bertrums
About one in fourteen children with acute myeloid leukaemia has a NUP98 fusion, and only about a third of them are alive five years later, against roughly two thirds of other children. It is four to five times commoner in children than in adults.
The same thing, for a scientist
The reference cohort is **160 NUP98-rearranged patients among 2,235 evaluable paediatric AML** from four Children's Oncology Group trials (CCG-2961, AAML03P1, AAML0531, AAML1031) — **7.2%** [E1]. The same paper contrasts **13 of 825 adults (1.6%)**, p < 0.001, with Beat AML contributing 0 of 440 [E1]. Outcome overall: **5-year OS 35% versus 64%** (p < 0.001), CR after course 1 **50% versus 78%**, DFS 27% versus 52% [E1]. Partner distribution: **NSD1 n=108, KDM5A n=32, and 20 'NUP98-X' cases spread across 13 different partners** [E1]. The partners behave differently in ways that matter clinically: **NUP98::NSD1 has 5-year OS 36%, EFS 17% and CR after course 1 of only 38%** (induction failure), while **NUP98::KDM5A has OS 30% and EFS 25% but a CR rate of 81%, indistinguishable from reference** (p = 0.729) — a relapse disease rather than an induction-failure disease [E1]. Ages differ too: NSD1 median 10.2 y, **KDM5A median 2.7 y** (p < 0.001), NUP98-X median 7.9 y and bimodal. Breakpoints cluster tightly: **~85% of NUP98 fusions break in exon 12 (39.7%) or exon 13 (44.9%)**, with exon 14 breakpoints almost exclusively NUP98::KDM5A [E1]. The most consequential finding for practice at the time: **97% of KDM5A and 95% of NUP98-X cases had been classified standard-risk** [E1].
Frequency, across sources
| population | NUP98-r frequency | source |
|---|---|---|
| paediatric AML, 4 COG trials | 7.2% (160/2,235) | 1 |
| paediatric AML, French series | 3.8% (22/574) | 2 |
| paediatric AML, multi-group NUP98::KDM5A only | 2.0% (47/2,393) | 4 |
| adult AML (same paper as the paediatric 7.2%) | 1.6% (13/825) | 1 |
The spread between 3.8% and 7.2% is large and is not resolved here. Plausible contributors: NUP98::NSD1 and NUP98::KDM5A are cryptic — invisible on G-banding — so a series relying on karyotype under-counts, whereas 85% (17/20) of the NUP98-X partners were visible on G-banding 1. A cohort screened by RNA sequencing will find more than one screened cytogenetically. The direction of that bias is predictable and it favours the higher number.
The paediatric-versus-adult contrast is the cleanest number in the field because it comes from within a single analysis 1, and it is the reason this beat belongs to a paediatric project.
Outcome by partner, and why pooling them is a mistake
5-year, from the COG cohort 1:
| group | n | OS | EFS | CR after course 1 | relapse risk |
|---|---|---|---|---|---|
| all NUP98-r | 160 | 35% | — | 50% | — |
| NUP98::NSD1 | 108 | 36% | 17% | 38% | 64% |
| NUP98::KDM5A | 32 | 30% | 25% | 81% (p = 0.729, NOT impaired) | 68% |
| NUP98-X (13 partners) | 20 | 35% (p = 0.009) | 35% (p = 0.333, NS) | — | 69% (NS) |
| reference | — | 64% | 47% | 78% | — |
Multivariable: NSD1 OS HR 1.463 (1.1–1.94, p = 0.009) and EFS HR 2.032 (1.59–2.59, p < 0.001); KDM5A OS HR 1.825 (1.13–2.96, p = 0.015) with EFS and relapse risk not significant 1.
The two commonest partners therefore fail in different ways: NSD1 patients largely do not achieve remission, while KDM5A patients do achieve remission and relapse from it. That distinction matters directly for this beat, because a menin inhibitor that produces differentiation rather than killing is a poor fit for induction failure and a better fit for relapse prevention.
And within NUP98::KDM5A, even the breakpoint matters: exon-13 junction cases (n=19) had OS 51% against 0% for exon-14 cases (n=12), p = 0.011 1. Twelve patients is small, but a 51-point survival difference by exon is not a subtlety.
The exception the authors themselves flag
Verbatim 1: NUP98 fusions "portend dismal overall survival, with the noteworthy exception of patients bearing abnormal chromosome 13".
Abnormal chromosome 13 is strongly linked to NUP98::KDM5A — present in 19 of 30 evaluable cases (63.3% versus 2.3%, p < 0.001), with del(13q) in 43.3% 1. So a cytogenetic feature that travels with one partner marks a better-outcome subgroup inside the worst-outcome fusion class. Nobody has explained why.
The authors' overall recommendation nonetheless runs the other way: NUP98 fusions should be reclassified as a high-risk subtype regardless of partner 1. This entry records both, because the recommendation is about risk stratification policy under uncertainty and the exception is about biology, and they are not in conflict.
The transplant-era counter-evidence
The 35% figure comes from cohorts largely treated in an earlier era. More recent transplant-heavy series report much better outcomes, and this entry does not treat 35% as the current expectation.
One single-centre paediatric cohort of 32 NUP98-rearranged patients, 29 transplanted, reports 3-year OS 81.2% (95% CI 65.9–96.5), EFS 62.8% and cumulative incidence of relapse 28.1% 5. That is not a small improvement over 35%; it is a different disease trajectory.
The obvious cautions apply and are not rhetorical: single centre, 32 patients, three-year rather than five-year follow-up, and a cohort in which 91% were transplanted, which is a selected population — a patient must survive to transplant and be fit for it. Selection of that kind can move a survival curve without any therapy being better.
The practical consequence for this beat is not that NUP98-rearranged AML is now curable, but that the comparator for a menin inhibitor is moving. A drug offering a 5.9-month median remission (trial-save-revumenib-nup98-subgroup) is a bridge-to-transplant proposition, and the value of that bridge depends on what transplant delivers — which is exactly the number that is changing.
| metric | frequency and survival |
|---|---|
| population | paediatric acute myeloid leukaemia |
| region | North America (COG trials), with European and Japanese comparators |
| year | 2023 |
| source | Children's Oncology Group CCG-2961, AAML03P1, AAML0531, AAML1031 |
| n evaluable paediatric | 2235 |
| n nup98 rearranged | 160 |
| frequency paediatric PCT | 7.2 |
| frequency adult PCT | 1.6 |
| n adult evaluable | 825 |
| paediatric vs adult p | <0.001 |
| os 5yr nup98r PCT | 35 |
| os 5yr reference PCT | 64 |
| cr after course1 nup98r PCT | 50 |
| cr after course1 reference PCT | 78 |
| dfs 5yr nup98r PCT | 27 |
| kdm5a breakpoint outcome | exon-13 junction OS 51% (n=19) vs exon-14 OS 0% (n=12), p=0.011 |
| chromosome 13 exception | abnormal chromosome 13 in 19/30 evaluable KDM5A cases (63.3% vs 2.3%, p<0.001); the authors name these patients as the noteworthy exception to dismal survival |
| risk misclassification at the time | 97% of KDM5A and 95% of NUP98-X had been classified standard-risk |
| authors recommendation | reclassify NUP98 fusions as high-risk regardless of partner |
| transplant era counterpoint | single-centre paediatric cohort n=32 (29 transplanted): 3-year OS 81.2% |
by partner — the numbers behind the chart
{
"NSD1": {
"n": 108,
"os_5yr_pct": 36,
"efs_5yr_pct": 17,
"cr_course1_pct": 38,
"relapse_risk_pct": 64,
"median_age_years": 10.2
},
"KDM5A": {
"n": 32,
"os_5yr_pct": 30,
"efs_5yr_pct": 25,
"cr_course1_pct": 81,
"cr_course1_p": 0.729,
"relapse_risk_pct": 68,
"median_age_years": 2.7
},
"NUP98-X": {
"n": 20,
"n_distinct_partners": 13,
"os_5yr_pct": 35,
"efs_5yr_pct": 35,
"efs_p": 0.333,
"median_age_years": 7.9
}
}breakpoint distribution
{
"exon_12_pct": 39.7,
"exon_13_pct": 44.9,
"exon_14": "almost exclusively NUP98::KDM5A (p<0.001)"
}- If modern transplant already delivers 3-year OS near 80% in selected paediatric NUP98-rearranged AML, a menin inhibitor with a 5.9-month median remission has to justify itself as a bridge rather than a therapy, and a trial powered against the historical 35% would be measuring against the wrong comparator.
- NUP98::NSD1 fails at induction (38% CR after course 1); a differentiating agent is a poor fit for induction failure, so the largest NUP98 subgroup may be the one least suited to the intervention this beat is arguing for.
- Partner-blind risk stratification put 97% of NUP98::KDM5A patients in the standard-risk arm. Partner-blind trial eligibility repeats the same error in the opposite direction, pooling an induction-failure disease with a relapse disease.
- Why does abnormal chromosome 13 rescue outcome inside the worst-outcome fusion class? It is the most striking unexplained observation in the landscape paper and nobody has followed it up.
- Why does the NUP98::KDM5A exon-14 breakpoint give 0% survival against 51% for exon-13? A 12-patient observation, but if real it means the breakpoint, not just the partner, should be reported.
- Is the gap between 3.8% and 7.2% entirely explained by cryptic fusions escaping karyotype-based screening? If so, the true frequency in unscreened populations is higher still and NUP98 is being under-diagnosed.
- Has the transplant era genuinely moved 5-year OS from 35% toward 80%, or is the modern cohort selected? A cooperative-group re-analysis restricted to recently treated patients would answer it and would set the bar any menin inhibitor has to clear.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Among 2,235 evaluable paediatric AML patients on four COG trials, 160 (7.2%) had a NUP98 rearrangement versus 13 of 825 adults (1.6%, p<0.001); 5-year OS was 35% versus 64%; partners were NSD1 n=108, KDM5A n=32 and 20 cases across 13 other partners; NUP98::NSD1 had CR after course 1 of 38% and EFS 17% while NUP98::KDM5A had a CR rate of 81% (p=0.729) and OS 30%; ~85% of breakpoints fall in NUP98 exon 12 or 13; and 97% of KDM5A and 95% of NUP98-X cases had been classified standard-risk.
Comprehensive molecular and clinical characterization of NUP98 fusions in pediatric acute myeloid leukemia · Bertrums EJM, et al. · Haematologica 108(8):2044-2058 2023What this evidence cannot show
Retrospective analysis of cooperative-group trial cohorts accrued over roughly two decades, so treatment eras differ within the cohort and the adult comparator groups (Beat AML, TCGA, SWOG) were assembled by different methods from the paediatric ones. The NUP98-X group is 20 patients across 13 partners, so every partner-level statement inside it rests on one to four cases. Subgroup survival differences were not pre-specified.
Evidence that argues against this (1)
- A recent single-centre paediatric cohort of 32 NUP98-rearranged patients, 29 of them transplanted, reports 3-year OS of 81.2%, far above the 35% five-year figure here. Either transplant in the modern era changes the disease, or the single-centre cohort is selected; either way the 35% should not be quoted as the current expectation without saying which era it describes. Zhang ZX et al. 2026
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
An independent French series found NUP98 rearranged in 3.8% of paediatric AML (22 of 574), forming a clinically and molecularly homogeneous group with a poor prognosis, and noted this frequency is similar to CBFB-MYH11.
NUP98 is rearranged in 3.8% of pediatric AML forming a clinical and molecular homogenous group with a poor prognosis · Struski S, et al. · Leukemia 31(3):565-572 2017What this evidence cannot show
3.8% is roughly half the COG figure. The discrepancy is most likely a detection-method difference, since NUP98::NSD1 and NUP98::KDM5A are cryptic on karyotype, but the two papers' screening strategies were not compared directly here, so the reconciliation is an inference.
Evidence that argues against this (1)
- This paper describes NUP98-rearranged AML as a homogeneous group, while the larger COG analysis finds the two commonest partners fail in opposite ways - induction failure for NSD1 versus relapse from an 81% CR rate for KDM5A. Homogeneity is a claim the larger dataset does not support. Bertrums EJM et al. 2023
- E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
NUP98-KDM5A occurs in 2.0% of paediatric AML overall (47 of 2,393 across COG, AIEOP, BFM, CPH, DCOG and LAME), at a median age of 3.2 years versus 9.4 years, with 5-year EFS 29.6 +/- 14.6% and OS 34.1 +/- 16.1%, independent for OS (HR 1.77, p=0.009) and relapse (HR 1.62, p=0.049) but NOT for EFS (HR 1.40, p=0.08).
The clinical and biological characteristics of NUP98-KDM5A in pediatric acute myeloid leukemia · Noort S, et al. · Haematologica 106(2):630-634 2021What this evidence cannot show
Published as a Letter, so methodological detail is compressed. The EFS confidence intervals are very wide (+/- 14.6% and +/- 16.1%), reflecting the small number of events, and the EFS hazard ratio does not reach significance - so NUP98-KDM5A is an independent predictor of death and relapse but not, in this analysis, of event-free survival, which is an internally awkward result.
Evidence that argues against this (1)
- Within paediatric acute megakaryoblastic leukaemia, the same fusion was reported NOT to be an independent prognostic factor, because the reference population is itself poor-risk. Whether NUP98-KDM5A is prognostic depends entirely on what it is being compared with. de Rooij JD et al. 2013
- E5moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper
A single-centre paediatric cohort of 32 NUP98-rearranged AML patients, 29 of whom underwent transplant, reported 3-year OS 81.2% (95% CI 65.9-96.5), EFS 62.8% and cumulative incidence of relapse 28.1%.
Clinicogenetic characteristics and outcomes of children with NUP98-rearranged acute myeloid leukaemia: A single-centre cohort study of 32 cases · Zhang ZX, et al. · British Journal of Haematology 2026What this evidence cannot show
Single centre, 32 patients, 3-year rather than 5-year follow-up, and 91% transplanted - a population selected for surviving to transplant and being fit for it. Selection of that kind can lift a survival curve without any therapy being better, and the confidence interval reaches 96.5%, which is implausible as a population estimate.
Evidence that argues against this (1)
- An adult multicentre transplant series found that pre-transplant NUP98-rearrangement status had NO prognostic significance at all, and that what predicted relapse was measurable residual disease one month AFTER transplant. If the post-transplant measurement is what matters, a single-centre cohort's excellent outcome may reflect its transplant programme rather than the fusion. Yuan X et al. 2025
- related to →NUP98::NSD1, t(5;11)(q35;p15.5)
108 of 160 NUP98-r cases; 5-year OS 36%, EFS 17%, CR after course 1 38%
- related to →NUP98::KDM5A (NUP98-JARID1A), t(11;12)(p15;p13)
32 of 160; CR rate 81% but relapse risk 68%; median age 2.7 y
comparator fusion class in the same paediatric AML cohorts
sets the comparator a 5.9-month remission has to beat
- related to →NUP98 — domain architecture and residue coordinates (with the 1817/920 numbering trap resolved)
85% of breakpoints in NUP98 exon 12 or 13
its exon-13 junction sits in the family's dominant breakpoint bin
much of the fusion's poor prognosis may belong to FLT3-ITD